The Noises Drivers Report to the Federal Government — and Which Ones Show Up Alongside Crashes and Fires

If you would rather have software listen first, every AI sound tool compared lists what each costs.

On 5 September 2026 we queried the National Highway Traffic Safety Administration's open complaints API for the 26 vehicles this site has guides for, across model years 2015 to 2022, pulled 121,288 distinct complaints, and counted how often drivers used each of 24 noise words in the free-text field. 12,607 contain at least one. Then we counted how many of those also carry the agency's crash or fire flag. The API is free and the script is published.

This is not the noise-price guide. Ten car noises ranked by repair cost sorts by the size of the bill, from an $11,990 knock down to a $511 hum. This page sorts by how many drivers wrote the word down on a federal form, then by how many of those reports mention a crash or a fire. A noise can be near the top here and cheap to fix, or absent here and ruinous: complaint frequency and repair cost are unrelated axes.

I built Pulscar — an AI tool that identifies car problems from a 30-second recording — after $6,000 of misdiagnosed repairs, starting with a $380 bill for an "engine knock" that was a loose heat shield and a $5 zip tie. Nobody lied to me: ninety seconds of listening bought a pattern-match to the expensive answer.

So here is the sentence that governs everything below. A complaint that mentions a knock and also carries crash: true does not show that the knock caused the crash. It shows that both facts appear in the same report. The language throughout is appears alongside, never causes.

How to read this list

Each item is a word, not a diagnosis. Item 2 is not "rod knock"; it is the string knock as typed by 1,420 drivers, covering rod bearings, detonation, exhaust leaks, suspension impacts and a few people describing somebody at their window.

One rule overrides everything: the rank counts words, not risk. The top word is pop, and its largest single cluster is a door hinge on a 2020 Ford Escape — a fact about people, not danger.

A second: a co-occurrence count is a floor, not an estimate. When we say 24 knock complaints also carry a crash or fire flag, exactly 24 documents contain both. We are not estimating how many knocks lead to fires. That number is not in this data.

How the count was made: the sample, the words, the limits

The sample was defined before anything was counted. We queried NHTSA's complaintsByVehicle endpoint for the 26 nameplates with a guide on this site, across model years 2015 through 2022: Honda Accord, Civic, CR-V, Pilot; Toyota Camry, Corolla, Highlander, RAV4, Tacoma; Ford F-150, Escape, Explorer, Fusion; Chevrolet Silverado 1500, Equinox, Malibu; GMC Sierra 1500; Ram 1500; Jeep Grand Cherokee, Wrangler; Nissan Altima, Rogue; Hyundai Sonata; Kia Sorento; Subaru Outback; Volkswagen Tiguan.

NHTSA splits some nameplates by body style and powertrain, so the F-150 alone needed twelve model strings from the agency's own list. That gave 768 requests, none of which failed, de-duplicated by ODI number. We excluded the Corolla Cross, F-150 Lightning, Explorer Police Interceptor, the 2500 and 3500 trucks and the Rogue Sport.

The words. Twenty-four whole-word, case-insensitive patterns: noise, pop, knock, click, rattle, grind, clunk, whine, tick, bang, squeak, screech, squeal, rumble, hum, thump, whistle, buzz, roar, chirp, hiss, groan, howl, grumble. They match the free-text summary, which the ODI dictionary calls CDESCR, "DESCRIPTION OF THE COMPLAINT". A complaint counts once per word. The umbrella word noise is reported separately, not as a ranked item.

The false friends, published rather than applied quietly. Before correction pop matched 3,041 complaints, most of the surplus being warnings that "popped up" and hoods that "popped open"; removing those took it to 1,663. knock matched the knock sensor, tick a ticking time bomb. Every pattern and removal count is in the sources log.

The correction that mattered most. The raw scan showed knock in 61 crash-flagged complaints. Reading the matches showed almost all were the word as a plain verb inside a crash narrative — knocked unconscious, knocked down a bush, knocked him over: collisions describing themselves, not engines. After removing the transitive uses of knock and bang, knock's crash co-occurrence is 8, not 61. The instrument was wrong in exactly the direction that would have flattered the article.

The fix costs something. ODI 11689260, a 2017 Honda Accord, fire-flagged, reads: "The rod knocked the starter loose while the vehicle was going 70 MPH, and caused a fire." That is a real rod-knock report, and our "knocked … loose" rule throws it out. A filter you bend for individual records is not a filter, so it stayed — but the loss is on the record.

What the crash and fire flags do not mean

They are answers on a form. The ODI dictionary defines field 7 as "WAS VEHICLE INVOLVED IN A CRASH, 'Y' OR 'N'" and field 9 as "WAS VEHICLE INVOLVED IN A FIRE 'Y' OR 'N'" — questions put to an owner filling in NHTSA's Vehicle Owner's Questionnaire, not the agency's determination.

The data is self-reported and unverified. NHTSA's own description says what the file is for: "Complaint information entered into NHTSA's Office of Defects Investigation vehicle owner's complaint database is used with other data sources to identify safety issues that warrant investigation and to determine if a safety-related defect trend exists." That is complaints as a signal that may justify looking, not a finding. We searched NHTSA's datasets page, reporting page, safety-issue search and flat-file documentation for an explicit sentence calling individual complaints unverified, and could not locate one, because nhtsa.gov returns HTTP 403 to automated requests. So, in our words: treat every report here as one person's unverified account.

A noise in a crash-flagged complaint does not mean the noise caused the crash. One document contains both. Often the arrow points the other way — several bang complaints here describe the airbag going off. Nothing separates those cases at scale, so we compute no rates, no percentages, no risk ratios. "Of 1,420 complaints mentioning a knock, 24 also carried a crash or fire flag" is defensible. "Knocks are X times more dangerous" is not a sentence this data can produce.

Nobody died in any of them. Across all 12,607 noise-word complaints the deaths field sums to zero, against 61 reported deaths in the full sample. Fatal complaints exist; they do not describe sounds.

Is your noise even in this vocabulary? The 5-minute filter

Engine speed or road speed? Reach the speed where the noise lives, shift to neutral, coast. Pitch holds while the engine drops: road speed — wheels, bearings, tires, driveline. Pitch falls with the engine: engine speed — belts, valvetrain, exhaust, timing, internals.

Continuous or triggered? A trigger is a component list. Only when braking narrows to about five parts; only at full lock, about six; only on a cold start, three, one of them normal.

Did anything arrive with it? The question this data answers best: smoke, a burning smell, a warning light, lost power, a changed pedal. The words that co-occur with fire flags almost always sit in a sentence containing smoke or flames.

Is it new, or has it always done that? A noise unchanged for two years is a different object from one that appeared last month, and this file is weighted toward the second.

Is your car in scope? Everything below comes from petrol, hybrid and diesel vehicles of model years 2015 to 2022, from 26 mainstream American nameplates. Drive an EV or a 2004 anything and the vocabulary applies, the concentrations do not.

The 15-minute protocol, before you spend anything

1. Check the oil, cold, on level ground. Level and colour. An engine two quarts down will knock. → items 2 and 8.
2. Idle it with the hood up. Note whether the noise changes as the gauge moves. → items 4 and 8.
3. Blip the throttle in neutral, wheels still. If the sound follows engine speed, it is on or in the engine. → items 2, 7 and 8.
4. Drive to the speed where it lives, then coast in neutral. Pitch unchanged means road speed. → items 5, 7 and 10.
5. Brake hard once from 30 mph, somewhere empty. Grinding, a pulsing pedal and a longer-travelling pedal are three different findings. → items 5 and 10.
6. Turn full lock both ways at walking pace. A rhythmic click here is a short list. → item 3.
7. Cross a speed bump slowly, windows down. A single clunk from one corner is suspension, not driveline. → item 6.
8. Record thirty seconds while it is happening, saying at the start what you are doing. → every item.
9. Run your VIN against open recalls. Free, and several clusters below sit on nameplates with a campaign. → items 2, 4, 8 and 9.
10. Write down the date and what changed. The most useful thing in a work order is a before and an after. → every item.

Find your situation

What actually determines your number

Seven factors decide what a complaint-file rank means for your car. None is the rank.

1. How many of your car exist. Volume tracks fleet size. The F-150 and Escape top the per-vehicle table with 9,727 and 9,747 complaints of all kinds because America bought millions.

2. Whether owners of your car are organised. A forum thread saying "report this to NHTSA" produces a spike, and that spike is behaviour, not a change in failure rate.

3. Whether the word is specific to one failure. whine is informative: 394 of its 669 complaints are one nameplate and 363 sit under POWER TRAIN. pop is not, covering a door hinge, a tire, an electrical arc and a fuel line.

4. What arrived with the noise. Smoke, a burning smell, a warning light or lost power move a sound from annoyance to stop-driving event.

5. Whether a federal campaign already names your problem. Several clusters below sit on vehicles with a recall covering the underlying failure, including one Kia remedy named the Knock Sensor Detection System. A VIN check can move the price to zero.

6. Whether the noise is changing. A rattle that lasted one second last winter and five now is a different report from one identical for three years. The file cannot see this; you can.

7. Model year, not calendar year. The pop cluster is 2020, the rattle cluster 2017 and 2018, the whine cluster 2019 to 2021. A neighbouring model year is often a different vehicle underneath.

The ten noises, ranked by how often drivers report them

Ranked by complaints containing each word, after the corrections above. The crash and fire counts mean both facts appear in the same report — nothing more.

1 · Pop — 1,663 complaints

The tell. One sharp report, not a rhythm, splitting three ways: body hardware when a door opens, an electrical arc or smell, something mechanical letting go.

What it actually is. Three unrelated things in one word. ODI 11701816, a 2020 Escape: "Drivers door check makes loud popping sounds , hard to open door ." ODI 11373646, a 2018 Wrangler with a fire flag: "I WAS DRIVING ON THE INTERSTATE HIGHWAY WHEN I HEARD A LOUD POP AND SAW AN ARC FLASH".

The cheap cause that gets skipped. Body hardware. The dominant cluster behind the top-ranked word in federal complaint data is a door check arm and hinge, tens of dollars, skipped because a pop sounds structural.

The confirmation test, free. Reproduce it with the ignition off: a pop with the car dead, only when a door or hatch moves, cannot be electrical or mechanical. If it needs the engine running, ask what arrived with it.

Where this does not apply. On an EV the mechanical family shrinks and the electrical grows. The Escape concentration is model year 2020; a 2019 or 2022 Escape is not the same door.

At the shop — say this: "The pop happens with the ignition off when the door moves — check the door check arm and hinge first. If it also happens with the engine running, inspect the fuel line and harness, and run my VIN against the open Escape fire recalls."

2 · Knock — 1,420 complaints

The tell. Heavy, hollow, rhythmic hammering tracking engine speed, loudest at idle, felt as much as heard. ODI 11678171, a 2018 Grand Cherokee with a fire flag, records "an abnormal knocking sound coming from the vehicle while depressing the accelerator pedal" before the RPM fell away.

What it actually is. Overwhelmingly the bottom end: a connecting-rod bearing losing its oil film and hammering the crank journal — hence the lopsided ENGINE share, and hence the two leading nameplates being those with the longest recall history for that failure.

The cheap cause that gets skipped. The oil level, then the octane. An engine two quarts low knocks; so does a loose heat shield; so does detonation on cheap fuel under load. None is the crankshaft, and all three are checked in a minute.

The confirmation test, free. If the knock follows engine speed in neutral with the wheels still, it is on or in the engine. Then listen from underneath while somebody revs: a shield or manifold leak gets louder there, a rod knock does not. More in the engine knocking guide.

Where this does not apply. An EV cannot make this noise. A diesel clatters at idle by design; the only question is whether it changed. On a hybrid, listen at the day's first start.

At the shop — say this: "Does the knock track RPM in neutral, and were oil level and pressure recorded first? Run my VIN — there are open rod-bearing campaigns on several 2015 to 2022 platforms."

A noise is evidence that exists only while it is happening. Record thirty seconds and Pulscar's AI returns the most likely cause, whether it is safe to drive, and the fair 2026 price — free, about ten minutes. The report arrives by email as a PDF, so use the address you actually read.

3 · Click — 1,218 complaints

The tell. A repeating tick-tick-tick with a mechanical edge, used for four unrelated things: a CV joint at full lock, a starter solenoid that will not crank, a parking pawl that did not engage, a chattering relay.

What it actually is. Whichever your context selects. ODI 11732629, a 2017 Civic crash-flagged with one injury: "an abnormal clicking sound, and the steering wheel locked up". ODI 11349078, a 2016 Sorento, also crash-flagged: "AN ABNORMAL CLICKING NOISE WAS HEARD. THE VEHICLE ROLLED BACK ON IT'S OWN". Two different failures, one word — which is why click carries the highest crash count here without saying anything about CV joints.

The cheap cause that gets skipped. The battery and its terminals. A rapid machine-gun click when you turn the key is a starting-circuit voltage problem far more often than a starter, and cleaning a terminal is free. Runner-up: a stone in a brake dust shield. Both in the clicking noise guide and clicking when starting.

The confirmation test, free. Sort by trigger, not by sound. Key turns, engine does not: electrical — see car will not start, clicking noise. Only at full lock while moving: CV joint. Once per wheel revolution: something touching a rotating part. Into park on a slope: the pawl, a rollaway risk several complaints describe.

Where this does not apply. An EV still clicks at full lock but has no starter solenoid. On a manual, the parking-pawl family does not exist.

At the shop — say this: "The click only happens at full lock while moving, so check the CV joints — or, if it happens when I park on a hill, inspect the parking pawl and shift cable, because the car has rolled after I put it in park."

Why does the same noise get reported thousands of times and still get misdiagnosed at the counter? Because a word is not a measurement. Nothing in the string "rattle" tells a technician whether the sound is a heat shield or a timing chain, and nothing in a complaint file separates them either, which is exactly why 1,193 people typed one word about at least four unrelated failures. The distance between the cheap cause and the expensive one is filled entirely by diagnostic work that not every shop performs, and the defence is a single sentence that never changes: show me the measurement. Chain stretch in millimetres. Pad thickness per corner. Fluid condition in writing. Play demonstrated by hand. Pulscar gives you an independent read on the sound first, so you walk in with a hypothesis instead of a symptom and a fear.

4 · Rattle — 1,193 complaints

The tell. Loose metal in a tin. What matters is not pitch but duration and direction of travel: two seconds on a cold start last winter, five now. ODI 11519578, a 2015 Sonata, fire-flagged, calls it "a rattling sound, which could best be described as tying cans to the car and dragging them along the road".

What it actually is. Given the ENGINE concentration, most often a timing chain slapping its guides while the hydraulic tensioner waits for oil pressure, or a converter substrate breaking up, or a heat shield.

The cheap cause that gets skipped. Oil, then a clamp. The tensioner is hydraulic, so low, degraded or wrong-viscosity oil produces a cold-start rattle on a sound engine. Under the car, a heat shield is bolted around the expensive part and sounds like it. Map in the rattling noise guide.

The confirmation test. Fresh oil and filter at the exact viscosity, then a week with the cold-start duration written down each morning. Shorter means that was the repair; unchanged means hardware — then ask for the chain stretch measured or shown to you. A tap test separates shields from the converter body.

Fix it yourself. A large hose clamp around a rattling heat shield, cold car, on stands, ten minutes. The honest limit: never under a car on a jack alone, never on a hot exhaust, and if the mounting tabs have rusted away a clamp is only a bandage. If the rattle is inside the engine, there is nothing here for hand tools.

Where this does not apply. An EV has no chain, exhaust or heat shields, so this collapses to trim and suspension. On a timing belt engine the failure is silent, which is why belts have an interval.

At the shop — say this: "Fresh oil at the exact spec first, then re-check how long the rattle lasts cold. If it is unchanged, I want the chain stretch measured and the heat shields tap-tested, before anyone quotes a chain and tensioner."

5 · Grind — 977 complaints

The tell. Metal on metal, either continuous under braking or a coarse rasp tracking wheel rotation. ODI 11166158, a 2016 Corolla crash-flagged with two injuries, records "A GRINDING NOISE THAT YOU COULD FEEL AND HEAR AS WELL AS WOULDN'T LET YOU PUSH ALL THE WAY DOWN ON BRAKES".

What it actually is. Under braking, the friction material is gone and you are stopping on steel, turning a pad job into pads plus rotors. Away from braking it is a bearing that lost its cage, or a starter drive that failed to disengage — ODI 11020757, a 2015 Accord, crash-flagged: "THE VEHICLE MADE A GRINDING SOUND AND FAILED TO START".

The cheap cause that gets skipped. Pads caught in time, and before them a stone or a rusted rotor face. A car left standing a week grows surface rust that scrapes for the first few stops and then goes — not a fault, but it gets quoted as one.

The confirmation test. Somewhere empty, brake firmly from 30 mph, ten times. Gone after two or three: surface rust. Still there: look through the wheel — any pad thinner than about three millimetres is finished. The grinding noise guide and grinding when braking separate the rest.

Where this does not apply. On a hybrid or EV, regenerative braking does most of the stopping, so pads last far longer and a grind is more likely surface rust or a slide seized from disuse than worn friction material.

At the shop — say this: "I want pad thickness in millimetres for each corner, written down, and the rotors measured against minimum, before we discuss calipers. If inner and outer pads on one wheel differ a lot, that means a seized slide."

6 · Clunk — 753 complaints

The tell. A single dull impact, not a rhythm. Two triggers separate the entry: going into gear or lifting off, versus going over a bump.

What it actually is. Into gear, driveline lash — a worn universal joint, a differential, a transmission mount, or software letting the driveline take up slack. Over bumps, suspension: stabilizer bar links first, then bushings, then ball joints. ODI 11726738, a 2017 Explorer crash-flagged with one injury, reports "the vehicle vibrated and made an abnormal clunking sound".

The cheap cause that gets skipped. Stabilizer bar links. The loudest cheap part in the suspension, they fail constantly on trucks, and their noise is routinely quoted as control arms or struts. Grab one and shake it. The rest is in the clunking noise guide.

The confirmation test. Walking pace over a speed bump, windows down, one corner at a time. A clunk only over bumps at one corner cannot be the transmission; a clunk only when selecting drive or reverse, standing still, cannot be suspension.

Where this does not apply. On an EV there is no gear-selection lash, so the into-gear family disappears while the over-bumps family stays. On a solid-axle truck the parts change names: track bar and steering damper, not control arm and link.

At the shop — say this: "The clunk only happens over bumps at the front left, so check the stabilizer link and bushings on that corner, and show me the play by hand before we talk about control arms."

7 · Whine — 669 complaints

The tell. A rising pitch tied to speed. ODI 11735774, a 2017 Highlander: "The transmission is whining when i coast."

What it actually is. The sharpest concentration in this dataset. Complainants name the mechanism and the paperwork themselves: ODI 11749177 describes "the UA80 8-speed transmission whine defect (T-SB-0160-18)" and a Toyota customer support programme. We attribute that index to the owner who typed it, not to Toyota; we did not open the bulletin.

The cheap cause that gets skipped. Fluid, then the calendar. Degraded fluid produces exactly the whine-and-shudder complaint that gets quoted as a replacement, at a fraction of the price of what it can rule out. The second cheap cause is that the whine is normal: electric power steering whines on lock by design, a hybrid drive unit because reduction gears do. More in the whining noise guide.

The confirmation test. At the speed where the whine is clearest, shift to neutral and coast. Pitch unchanged means road speed: driveline or wheels. Pitch falling with the engine means an accessory or a belt.

Where this does not apply. On an EV a rising whine is the motor and reduction gear, and normal. On a hybrid with an eCVT there is no valve body, clutch pack or torque converter, so the expensive branch does not exist. On a manual, a whine in one gear only is that gear's bearings.

At the shop — say this: "The whine holds its pitch when I coast in neutral, so it is road speed. I want fluid condition in writing, and my VIN checked against any transmission customer support programme, before anyone quotes a transmission."

8 · Tick — 619 complaints

The tell. A light, fast, metallic tap following engine speed, often louder under load. It is the word most likely to be normal and the one owners most often attach to a named defect: ODI 11539852, a 2017 Ram 1500, "wants to add the Hemi tick defect as well."

What it actually is. Valvetrain, injectors or exhaust. A direct-injection engine ticks at idle because the injectors are loud, and that tick is permanent and harmless. An exhaust manifold leak ticks loudest cold and quietens as the metal expands. A collapsed lifter ticks regardless of temperature.

The cheap cause that gets skipped. Oil, again, then a manifold gasket. Ticking is the classic low-oil-pressure symptom, and the second-cheapest answer is a leaking exhaust manifold gasket, not anything internal. The third is that it always did that. The ticking noise guide works through the rest.

The confirmation test. Cold start, then five minutes. A tick that fades as the engine warms is very likely an exhaust leak, the metal having expanded to close the gap. A tick that appears or worsens hot points at oil pressure. A tick identical hot and cold on a direct-injection engine is the design.

Fix it yourself. Check and correct the oil, then listen again next morning. The honest limit: finding an exhaust leak by hand means feeling for hot gas around a manifold on a running engine, which is how people burn themselves, and the online tricks involve either a scalding surface or flammable gas near hot metal. If the oil is right and the tick remains, this stops being a driveway job.

Where this does not apply. No valvetrain, injectors or exhaust on an EV, so this entry does not exist there. On a diesel the whole engine ticks by design; the only question is whether it changed.

At the shop — say this: "Oil level and pressure recorded first. Then tell me whether the tick fades as it warms — if it does, check the exhaust manifold and gasket before anyone opens the valve cover."

How do you tell a noise that warns you from a noise that is the failure itself? In this dataset the difference is almost never in the sound and almost always in what came with it. The fire-flagged complaints share a pattern that has nothing to do with pitch: the owner writes down smoke, a burning smell, a warning light or a sudden loss of power in the same sentence as the noise. Read a hundred of them and the sound stops looking like the signal at all. That is the practical rule this scan supports and the ranking does not: a noise on its own buys you time to run the cheap tests above, and a noise with heat, smoke or lost power buys you none. Pulscar is built for the first category, where a considered answer beats a guess made in a car park.

9 · Bang — 509 complaints

The tell. One loud report, usually with something else in the same second. ODI 11307999, a 2018 Wrangler, fire-flagged: "A LOUD BANG WAS HEARD … METAL CHUNKS AND PIECES COULD BE HEARD FALLING UNDERNEATH THE VEHICLE". ODI 11557451, a 2022 Ram 1500 crash-flagged with one injury: "I heard what sounded like an explosion/ loud bang" — a lower control arm had snapped.

What it actually is. Not one thing, and this is where the co-occurrence caveat matters most. Some bangs are a component failing: a control arm, a clutch pressure plate, a driveshaft. Some are a tire. Some are an airbag deploying after a collision, which is why bang carries 19 reported injuries on only 509 complaints. The word appears alongside crashes partly because crashes make this sound.

The cheap cause that gets skipped. Nothing here is cheap and nothing here is wait-and-see. The closest thing is an exhaust backfire on a rich mixture or a tire that struck debris. Unlike every other entry, the move is to stop and look, not to book a diagnosis for next week.

The confirmation test. Stop somewhere safe, get out, look underneath from a distance. Smoke, a fluid trail, a hanging component or a burning smell means do not drive it. Otherwise note the exact conditions and get the suspension and driveline inspected within days.

Where this does not apply. On a manual-transmission Wrangler or Gladiator of these years, a bang with a burning smell has a named federal campaign behind one of its causes, so the VIN check comes first. On an EV the exhaust and clutch families vanish; suspension and tires remain.

At the shop — say this: "There was a single loud bang and then a burning smell — inspect the underside for a fractured component and a fluid leak before the car is driven again, and run my VIN against open recalls."

10 · Squeak — 238 complaints

The tell. A high, thin, intermittent chirp, usually from a wheel under light braking or over bumps.

What it actually is. Very often the car doing its job. Brake wear indicators are metal tabs designed to squeal when a pad gets thin — a warning working correctly: pads soon, not pads now. The rest is dry bushings, a worn belt, or a brake component binding, as in ODI 11121599, a 2018 Wrangler, fire-flagged: "THE BRAKES MADE SQUEAKING NOISES WHEN THE BRAKE PEDAL WAS DEPRESSED. THE CONTACT NOTICED A BURNING ODOR AND SMOKE COMING FROM THE BRAKES."

The cheap cause that gets skipped. Dry hardware. Bushings, sway-bar mounts and brake slide pins squeak when the lubricant is gone, and greasing slides is a fraction of a brake job. A squeak that vanishes when you spray water on the belt is a belt or tensioner, not an engine. The squeaking guide and squeaking when turning split those.

The confirmation test. Coast in neutral without touching the pedal at the speed where it appears: that separates speed-related from brake-related. Then look at the pads — plenty left means hardware and lubricant; a thin pad means the wear indicator, telling the truth.

Where this does not apply. On a hybrid or EV regeneration means pads may be barely worn at high mileage, making a squeak far more likely dry slides or rust than worn friction material.

At the shop — say this: "Tell me whether this is the wear indicator or dry hardware — I want pad thickness in millimetres, and if there is pad left, I want the slides cleaned and greased rather than a pad set."

The words that just missed the top ten

A cutoff you cannot see is a cutoff you cannot check. Screech 220, with 10 crash flags and 69 under service brakes. Squeal 168, also brake-heavy at 80. Rumble 166. Hum 159. Thump 148. Whistle 126. Buzz 113. Roar 77, with 10 crash flags on a small base. Chirp 70. Hiss 59, 5 with a fire flag. Groan 33. Howl 24. Grumble 2.

The interesting one is hum, the noise most associated with a failing wheel bearing, at a tenth of pop. That gap says nothing about bearings and everything about what people file complaints about: a hum is gradual and gets fixed, a door that pops is infuriating every day. The humming noise guide covers what this file barely notices.

The price ladder: read it from the bottom up

Complaint data carries no repair costs — NHTSA does not ask. Every figure below is carried over from the noise cost guide, re-checked in September 2026, and marked snippet-sourced where the estimator page refuses automated fetching. Shop estimates, not our measurements.

Read it from the bottom up, because that is the direction the money runs: the top four lines are free, and between them they resolve or de-risk much of the list.

Your number, by what you drive

Two rows there invert the guide rather than adjusting it. EV owners should read the ranking as a list of things that mostly cannot happen to them — no reciprocating assembly, no timing chain, no exhaust, no valvetrain — leaving click, clunk, squeak and grind. 2020 Ford Escape owners should read item 1 backwards: the number-one noise word in federal complaint data is, for their car, most likely a door hinge, and treating it as an engine symptom is how a small repair becomes a large invoice.

What happens at the shop, checkpoint by checkpoint

Checkpoint 1 — the reproduction. "Did the car make the noise for you, and under what conditions?" A noise never reproduced was never diagnosed; if they could not, offer your recording.

Checkpoint 2 — engine speed or road speed. "Is this tied to engine RPM or to wheel rotation?" If the paperwork implies one and your coast test showed the other, resolve that before a part is ordered.

Checkpoint 3 — the diagnostic fee and the credit. "What is the charge, and does it come off the repair?" Ask before the car goes on the lift. The snippet-sourced national range for diagnostic labour is $122 to $233.

Checkpoint 4 — the cheap cause, named and ruled out. "Which cheap cause did you eliminate, and how?" Oil before timing parts. Links before control arms. Tires before bearings. Hinges before body structure. Fluid before a transmission.

Checkpoint 5 — the measurement, not the adjective. "Can you show me?" Millimetres of pad, rotor against minimum, chain stretch measured, play demonstrated by hand, a tap test while you watch.

Checkpoint 6 — the ceiling and the campaign. "Call me before anything over this amount — and has my VIN been run for open campaigns?" If it goes wrong, how to dispute a repair bill covers what a written estimate is worth.

Three diagnostic traps that cost the most

Trap 1: the complaint count read as a danger ranking. The situation: you see that pop tops the list with 60 fire-flagged complaints, and your car pops. What the shop says: nothing yet — this trap happens before you reach a shop, and it changes what you ask for. What is really happening: the largest cluster behind pop is a door check arm on a 2020 Escape. Rank measures how many people typed a word, not danger, cost or fleet-wide frequency. Price versus bill: a door check arm and hinge against an engine-bay teardown you asked for because a table frightened you. Your protective question: "Can we reproduce my pop with the ignition off, opening the door, before we open the hood?"

Trap 2: the noise in the crash report read as the cause of the crash. The situation: your car makes a noise that appears in crash-flagged complaints, and you conclude a crash is coming. What the shop says: "If it's making that noise I wouldn't drive it" — sometimes true, sometimes a sentence that sells a tow. What is really happening: co-occurrence is not causation, and here the arrow frequently points the other way, because collisions are loud. Our own scan shows how easily this misleads: before correction knock appeared in 61 crash-flagged complaints; after removing "knocked unconscious" and "knocked down", 8. Price versus bill: a $0 confirmation test tonight against an emergency decision made on a misread table. Your protective question: "Is this noise a warning before a failure, or the sound of a failure that already happened — and how would you tell on my car?"

Trap 3: the cold-start rattle that becomes a timing chain. The situation: a few seconds of rattle on the first start of the day, and an irregular oil-change history. What the shop says: "Chain, tensioner and guides. Call it two thousand." What is really happening: the tensioner is hydraulic. Old, low or wrong-viscosity oil produces this rattle on a healthy chain, and the ENGINE-heavy rattle cluster contains many cars whose real problem was service history. Price versus bill: an oil change at a snippet-sourced $158 to $187 nationally, or $123 to $153 on a Corolla, against $1,461 to $1,852 for the chain job. Your protective question: "Fresh oil at the exact specification first, then re-check the duration. If it is unchanged, show me the chain stretch measured rather than inferred."

Three quotes, decoded

These are composite scenarios, not individual customers — the arithmetic is typical of the pattern each noise produces. We do not publish invented testimonials, and every figure comes from this guide or the cost guide.

Scenario 1 — the door that becomes an engine. A 2020 Escape pops when the driver's door opens, and the owner arrives having read that pop is the most-reported noise word in federal complaints. Ask for the ignition-off reproduction test first and the branch closes: a pop that exists only when a door moves, with the car dead, cannot be a fuel line. Ask for it after an engine-bay inspection has been authorised, and you have bought the inspection. Lesson: the order of the tests decides the bill, not the noise.

Scenario 2 — the whine that can be fluid. A whine rising with speed on a transmission with no fluid service in its history, quoted against a replacement of $6,165 to $6,685. Ask for the neutral-coast test and the fluid condition in writing, and two cheaper branches open: a fluid service at $243 to $313, and a VIN check against any customer support programme for that transmission. Lesson: ask what the noise does in neutral before anyone prices a gearbox.

Scenario 3 — the squeak that was the car telling you something. A squeak from a front wheel under light braking, quoted as pads, rotors and a caliper. Ask for pad thickness per corner in millimetres and the arithmetic decides: plenty of material means dry slides, greased for a fraction of a brake job; a thin pad means the wear indicator did its job and pads at $265 to $322 are honest; a large inner-to-outer difference on one wheel means a seized slide, and the $890 to $960 caliper is honest too. Lesson: a measurement on the invoice turns three prices into one correct one.

Should you buy a car whose complaint file is full of noise words?

Start with what complaint volume measures. Owners motivated enough to fill in a federal form, driven by three things before reliability: how many of the car exist, how organised its owners are, how annoying the fault is to live with. The F-150 and Escape top our per-vehicle table with 9,727 and 9,747 complaints of every kind, and they are two of the best-selling vehicles in the United States. Reading that as a reliability ranking is a category error.

Then use the file the way it is genuinely useful: for concentrations, not totals. A word piling into two model years on one platform tells you far more than a total. Three from this scan:

• 394 of 669 whine complaints are Toyota Highlanders, clustering in 2019, 2020 and 2021 with 363 filed under POWER TRAIN.
• 465 of 1,193 rattle complaints are Ford F-150s, weighted to 2017 and 2018, with 578 of all rattles filed under ENGINE rather than STRUCTURE.
• 454 knock complaints fall on 2015-2022 Kia Sorento and Hyundai Sonata, platforms with a documented recall history for connecting-rod failures and a remedy that updates the Knock Sensor Detection System.

Then check paperwork rather than vibe. Three free steps beat any amount of forum reading: run the VIN against open recalls; look up that model year on NHTSA's complaint search and read twenty complaints rather than counting them; and drive the car through the protocol above, particularly the neutral coast and a cold start. The cold start matters most: a seller who warms the car first has removed the most informative thirty seconds of the day.

And be clear about what a clean file does not prove. Low counts on a low-volume model mean almost nothing. Complaints also lag: a 2022 vehicle has had less time to accumulate them than a 2015 one, visible in our data as a decline across newer model years for almost every word. The 26 guides linked above go year by year, and how to find an honest mechanic covers a pre-purchase inspection.

Which noise should you actually worry about, given that this ranking deliberately refuses to rank danger? The sound matters far less than its company. In this dataset the fire-flagged complaints are overwhelmingly the ones where an owner wrote smoke, a burning smell, a warning light or a sudden loss of power in the same sentence as the noise, while the rest of the file is full of sounds people lived with for months without consequence. So the rule that survives 121,288 real reports is not about pitch or rhythm: a noise on its own buys you time to run the cheap tests above, and a noise with heat, smoke or lost power buys you none. Pulscar exists for the first category, where thirty seconds of audio and a considered answer beat a ninety-second pattern-match at a counter.

Your situation right now: four playbooks

"I have a quote in my hand." Find your word above, read its cheap-cause paragraph, say its script at the counter, and ask whether the noise was reproduced. Do not authorise anything over a few hundred dollars in the conversation where you first heard the number.

"The noise is new and the car feels normal otherwise." Run the fifteen-minute protocol in order; it costs nothing and narrows ten words to one or two families.

"There is a noise and also smoke, a smell, a warning light or lost power." Stop. That combination, not the sound, is what the fire-flagged complaints have in common. Pull over, get out, and have it recovered rather than driven.

"I want to know whether my car has a known problem." Run the VIN against open recalls, then read twenty complaints for your exact model year rather than counting them.

After the repair: verify before you celebrate

Re-record from the same position. Thirty seconds before and thirty after is the cheapest proof of repair there is, and the strongest position if the noise returns inside a parts warranty. The recording protocol makes the two files comparable.

Verify in the condition that produced it. A cold-start rattle on the next cold morning, not a warm restart in the workshop yard. A brake noise at the speed where it lived.

Keep the invoice with part numbers, and ask for the old parts back. A grooved rotor, an unevenly worn pad or a stretched chain either tell the story you were told or they do not.

File a complaint if the fault is unresolved. It is the mechanism that produced every number on this page.

Prevent the repeats. Oil at the manufacturer's interval addresses items 2, 4 and 8; transmission fluid on a real interval rather than a "lifetime" one addresses item 7.

Your action plan: 10 minutes, today, this week

Next 10 minutes — free:

1. Check the oil level and colour.
2. Blip the throttle in neutral and note whether your noise follows the engine.
3. Write down exactly when it appears: cold start, braking, turning, over bumps, at 40 mph, opening a door.

Today:

4. Run the neutral-coast test, and the steering-load test if it is a wheel noise. They settle items 5, 7 and much of 10.
5. Record thirty seconds properly and run it through a free sound check.
6. Call a dealer with your VIN and ask what campaigns are open.

This week:

7. Book a diagnosis rather than a repair, with your item's confirmation test written on the work order.
8. Get the measurement — millimetres, play, fluid condition, chain stretch, tap test — before approving any part.
9. If the fault is real and unresolved, file your own complaint.

Sources and documents

Primary documents only. Every complaint is quoted verbatim with its ODI number, retrievable from the same open endpoint.

1. NHTSA Office of Defects Investigation complaints API — queried 5 September 2026: 26 nameplates, model years 2015-2022, 768 requests, 121,288 distinct complaints by ODI number, 12,607 with a noise word. Endpoint, example query
2. ODI complaints flat-file data dictionary, updated 30 April 2026. Field 7 CRASH: "WAS VEHICLE INVOLVED IN A CRASH, 'Y' OR 'N'". Field 9 FIRE: "WAS VEHICLE INVOLVED IN A FIRE 'Y' OR 'N'". Field 20 CDESCR: "DESCRIPTION OF THE COMPLAINT". Field 21 lists "VOQ = NHTSA VEHICLE OWNERS QUESTIONNAIRE". CMPL.txt
3. NHTSA's description of the dataset: "Complaint information entered into NHTSA's Office of Defects Investigation vehicle owner's complaint database is used with other data sources to identify safety issues that warrant investigation and to determine if a safety-related defect trend exists." From the Internet Archive capture of 28 August 2026, because www.nhtsa.gov returns HTTP 403 to automated requests. NHTSA Datasets and APIs, archived
4. NHTSA recall 25V274 — GM, 597,571 units, 2021-2024 6.2L V8 trucks and SUVs: "The connecting rod and/or crankshaft engine components may have manufacturing defects that can lead to engine damage and engine failure." Recall record
5. NHTSA recall 23V751 — Honda, 248,999 units including 2016 and 2018-2019 Pilot: "the connecting rod bearing in the engine may wear and seize, damaging the engine." Campaigns XG1 and GG0. Recall record
6. NHTSA recall 20V750 — Kia, 294,756 units: "An engine compartment fire can occur while driving." Remedy: "Kia dealers will update the Knock Sensor Detection System (KSDS) software, free of charge." Recall record
7. NHTSA recall 18V907 — Kia, 2012-2017 Sorento among others, on vehicles previously repaired under 17V-224 or the KSDS campaign: "A fuel leak in the presence of an ignition source can increase the risk of a fire." Recall record
8. NHTSA recalls 21V986 and 22V623 — Ford, 2021-2022 F-150: "Underbody heat and noise insulators may loosen and contact the aluminum driveshaft, which could damage the driveshaft and cause it to fracture." Recall record
9. NHTSA recall 22V191 — Ford, 2020-2022 Escape 1.5L: "The engine oil separator housing may crack and develop an oil leak," with a fire consequence. Companions: 22V859, 23V380, 25V165, 25V345, 25V467. Recall record
10. NHTSA recall 24V635 — Ford, 2021-2022 F-150 and others, 2.7L and 3.0L Nano EcoBoost: "The engine intake valves may break while driving, which can result in engine failure and a loss of drive power." Recall record
11. NHTSA recalls 20V124, 21V028 and 23V116 — FCA, 2018-2023 Wrangler and 2020-2023 Gladiator, manual gearbox: "The clutch pressure plate may overheat and fracture." Recall record
12. NHTSA recall 19V645 — GM, 3,456,719 units, brake vacuum pump: "the brake assist decreases, increasing braking effort." Listed only because ODI 11171074 quotes a manufacturer notice mentioning a ticking noise; we did not locate that notice. Recall record

What we could not verify. No explicit NHTSA sentence calling individual complaints unverified: nhtsa.gov blocks automated requests, and no such wording appears in the flat-file documentation, the data.gov entry or the archived pages we reached. So we say it ourselves: this data is self-reported and unverified. We did not open Toyota bulletin T-SB-0160-18 or the ZJC programme document — both indices come from complainants' own text, ODI 11749177 and ODI 11739802. We did not check for an NHTSA investigation into the 2020 Escape door cluster and do not claim one exists. No price here comes from complaint data, because NHTSA does not collect repair costs.

What to read next

Start with the complete noise index, then the deep dives on knocking, rattling, clicking, clunking, grinding, ticking, whining, humming and squeaking. Money view: noises ranked by repair cost. Sound against scanner: how the two compare. Free repairs in your VIN: warranty extensions owners do not know about.

How these numbers were built: one script against NHTSA's public complaints API, run on 5 September 2026 over a sample declared before counting; the full method, word list and exclusions are in the sections above and the sources log. No rate, share or risk ratio is computed anywhere. Repair prices come from the linked cost guide and vary by engine, parts choice, access and local labour rates.

Prices verified: September 2026 · Technical documents verified: September 2026 · Reviewed quarterly

Ran the scan and got a different number? Email [email protected] with your query and your counts.