Accuracy and standards
Timing your own watch: measuring and regulating
On this page (8 sections)
Almost every complaint about watch accuracy is based on bad measurement. Somebody compares a watch against a phone one morning, sees it is eleven seconds out, and concludes the watch is running fast. What they have measured is a week's drift plus an unknown starting error plus their own reaction time, divided by nothing.
Measuring a mechanical watch properly is not difficult, but it takes seven days and a little discipline. This guide covers the wrist test, the positional test, what a timing machine measures, and the honest boundary between what regulation can fix and what needs a service.
The seven-day wrist test
This is the test that matters, because it measures the watch doing the job you bought it for: on a wrist, in your temperatures, in your positions, wound the way you wind it.
- Choose a reference. Use a clock synchronised over NTP, which is what a phone or computer clock is when set automatically. Not a second wristwatch, a microwave, or a car dashboard.
- Wind the watch fully if it is manual, or wear it long enough to be fully wound if it is automatic. Starting at low wind measures the weak end of the mainspring rather than the watch.
- Set it precisely. Pull the crown to hack the seconds if the movement allows it, set the hands slightly ahead, and push the crown in exactly as the reference passes that time. If it does not hack, set it as close as you can and record the offset as your day-zero figure rather than pretending it is zero. Our guide to setting and adjusting your watch covers the method.
- Record the deviation at the same time each day, to the nearest second, for seven days. The same time keeps the wear-and-rest cycle constant. Write the raw deviation from the reference, not the change since yesterday.
- Note the pattern alongside it: hours worn, whether the day was active or sedentary, and the overnight resting position. This column is what turns a number into an explanation.
- Compute the mean daily rate: total deviation on day seven, divided by seven. Plus 31 seconds after a week is about plus 4.4 seconds a day.
- Repeat if the daily figures are wildly inconsistent. Plus 2, plus 9, minus 1, plus 7 across four days is telling you something about winding state or position that the mean will conceal.
One day of data is not a rate. A single reading contains your own error at both ends, easily a second or two, plus the genuine day-to-day variation caused by how much you moved and how the watch sat overnight. On a watch running at plus 3 seconds a day, that noise swamps the signal. Seven days divides your reading error by seven and averages the rest out.
The positional test
The wrist test tells you how the watch performs. The positional test tells you why.
A mechanical watch keeps a different rate depending on its orientation, because gravity acts differently on the balance and the pivots rest on different surfaces. Six positions are conventionally tested: dial up, dial down, crown up, crown down, crown left and crown right.
The method is simple and slow. Wind the watch fully, leave it in one position for 24 hours, record the deviation, then reset and repeat. Six days of work, re-winding at each stage so you compare like with like.
What you are looking for is not any single figure but the delta, the spread between the fastest position and the slowest. A watch running at plus 2 in every position is excellently adjusted. A watch running at minus 8 dial up and plus 14 crown down has a large delta and behaves unpredictably, because your daily rate then depends on how you happened to move that day. Delta is the number a watchmaker cares about, and the one a regulator arm cannot fix.
What a timegrapher is
A timegrapher, or timing machine, does in thirty seconds what the wrist test does in a week, and answers a different set of questions. It is a microphone in a clamp plus some software.
The clamp holds a piezoelectric sensor against the watch case. Every beat of the escapement produces a characteristic three-part sound: the unlocking of the pallet stone, the impulse, and the drop as the next tooth lands, all described in our guide to escapements. The machine times the intervals between those sounds far more precisely than you can time hands against a clock, and derives three numbers.
Rate, in seconds per day, is calculated from how far the actual beat interval differs from the movement's nominal frequency. It is an instantaneous figure, extrapolated, and not the same thing as the mean daily rate you get from wearing the watch. The two frequently disagree, because the machine holds the watch in one position at one state of wind.
Amplitude, in degrees, is how far the balance swings from its rest position in each direction. It is the best single health indicator in the movement, because it falls when energy is lost to friction, thickened oil, a tired mainspring or a fault, and it falls long before the rate misbehaves. A very high figure is its own warning: above roughly 320 degrees the impulse pin can strike the back of the pallet fork, which is called knocking and shows up as an erratic rate.
Beat error, in milliseconds, is the asymmetry between the tick and the tock. If the balance's rest point is not exactly aligned with the pallet fork's center, one half of the cycle is slightly longer than the other. The machine reports the difference.
Lift angle, and why it matters
Amplitude is not measured directly. The machine cannot see the balance. It infers amplitude from the timing of the escapement sounds, and to do that it needs to know the lift angle: the angular distance the balance travels while it is in contact with the escapement, from unlocking to the end of impulse.
Lift angle is a fixed property of a given movement, published by its manufacturer, and typically falls somewhere in the region of 40 to 55 degrees. Most timing machines default to 52 degrees, which is right for a good many calibers and wrong for plenty of others.
Get it wrong and the amplitude reading is wrong, roughly in proportion. Enter 52 degrees for a movement whose true lift angle is 44 and the machine will report an amplitude comfortably above the real one, which can turn a movement that needs a service into a movement that looks healthy. Rate and beat error are unaffected. If you take one thing from this section: look up the lift angle for your specific caliber before believing any amplitude number.
What good numbers look like
| Reading | Healthy range | What a poor figure suggests |
|---|---|---|
| Amplitude, dial up, full wind | Roughly 270 to 310 degrees | Below about 250 suggests old oil, a weak mainspring or wear |
| Amplitude, vertical, full wind | Typically 20 to 50 degrees lower than horizontal | A much larger drop suggests pivot or escapement problems |
| Beat error | Under about 0.5 ms is good, under 1.0 ms usually acceptable | Above roughly 1.0 ms can cause poor self-starting at low wind |
| Rate, single position | Within a few seconds per day of the target | A large figure alone is often just regulation |
| Delta across positions | Smaller is better; a well-adjusted movement is tight | A wide spread means adjustment or wear, not regulation |
| Trace lines | Two thin, straight, parallel lines | See below |
Amplitude falls naturally as the mainspring runs down, so all of the above assume a fully wound watch. A drop of 40 to 50 degrees between full wind and 24 hours later is ordinary.
Reading the trace
The timegrapher draws each beat as a dot, and consecutive beats form lines that scroll across the screen.
- Two straight parallel lines is what a healthy watch looks like. Their slope is the rate: up means gaining, down means losing, horizontal means spot on.
- The vertical gap between them is beat error made visible. Perfectly in beat, the two lines merge into one.
- Thick or fuzzy lines usually mean noise in the escapement: dirt, worn pivots, or a badly fitting clamp.
- A regular wavy pattern, repeating with the rotation of a wheel, points at something bent or out of true, such as a distorted hairspring or an eccentric escape wheel.
- Scattered dots with no line mean the machine cannot find a consistent beat: a fault, a wrong frequency setting, or a watch that is simply not running properly.
Do not over-read the trace. Its value at home is in the obvious cases, healthy or clearly not. Distinguishing the several possible causes of a fuzzy line is bench work.
The nightstand trick
Because rate varies with position, the eight or so hours a watch spends off the wrist each night are a free adjustment you can make without touching the movement.
Suppose a watch gains about 4 seconds a day in wear, and your positional test showed it runs at roughly minus 8 seconds a day sitting crown up. Sixteen hours on the wrist is two thirds of a day at plus 4, so about plus 2.7 seconds. Eight hours resting crown up is one third of a day at minus 8, so about minus 2.7 seconds. Over 24 hours the watch is close to zero.
The arithmetic is that simple: multiply each rate by the fraction of the day spent in that state and add. It only works if you have done the positional test first, so you know which position slows the watch and by how much.
Regulate, or service?
Regulation changes the effective length of the hairspring, which changes the rate. It cannot add energy, cannot reduce friction and cannot correct a spread between positions. So the decision comes down to what your measurements show.
| Symptom | Likely cause | What it needs |
|---|---|---|
| Consistent gain or loss, healthy amplitude, small delta | Rate simply set away from zero | Regulation |
| Suddenly gaining minutes a day | Magnetism, coils sticking together | Demagnetising, covered in our magnetism guide |
| Low amplitude at full wind | Degraded oil, weak mainspring, friction | Service |
| Amplitude collapses in vertical positions | Pivot wear, poor poise, escapement fault | Bench diagnosis, usually service |
| Large delta between positions | Poise error, worn pivots, hairspring not flat or centered | Adjustment by a watchmaker |
| Stops overnight but runs on the wrist | Low amplitude that survives motion but not stillness | Service |
| High beat error, poor starting | Hairspring collet not aligned with the fork | Beat correction at the bench |
| Rate fine when worn, wild when rested | Ordinary positional variation | Nothing, or the nightstand trick |
| Erratic day-to-day rate on an automatic | Insufficient winding from a sedentary day | Wear it more, or hand-wind |
The general rule: rate problems are regulation, energy problems are servicing. Our guide to servicing and care covers what a full service involves and how often it is genuinely needed.
Regulating it yourself, and the risks
Many movements can be regulated by the owner. On a traditional regulator, a small arm on the balance cock carries two curb pins that grip the hairspring; moving the arm toward the plus mark shortens the working length of the spring and speeds the watch up. On some movements the arm has a fine adjustment screw, which is far easier to use precisely.
The realistic assessment is this. The movement itself is a light nudge with a fine tool, and people do it successfully. The risks are that you touch the hairspring, which is a few hundredths of a millimeter thick and permanently ruined by a bend or a fingerprint; that you slip and mark the balance cock; and that you open a caseback without the right tools and damage the gasket, compromising water resistance. Free-sprung balances, which have no regulator arm and are adjusted by moving weights on the balance rim, are not owner-adjustable in any practical sense.
Beat error is harder again. On most movements, correcting it means rotating the hairspring collet on the balance staff so the balance's rest position lines up with the pallet fork, which requires removing the balance, the right collet tool and confidence that is earned rather than read about. Some modern movements have a movable stud carrier that makes it far safer, but even then a timing machine is needed to see what you are doing.
If you are not doing it yourself, ask for something specific. "Please regulate it" invites a single-position adjustment. Better: state your measured mean daily rate and how you wear the watch, ask for the rate to be set close to zero in the positions you actually use, for beat error above about 0.5 ms to be corrected, and for the positional delta to be reported back to you. If amplitude is low, ask whether regulation is worth doing at all before a service.
What accuracy is realistic
A modern, healthy, well-adjusted mechanical movement usually settles within about 5 seconds a day of zero in real wear. A chronometer-certified movement is tested to a mean daily rate of minus 4 to plus 6 seconds a day under a defined protocol, which our guide to chronometer standards explains in full. A well-regulated uncertified movement often matches that in practice, because the certification tests a movement in a laboratory rather than a watch on your arm.
What is not realistic is quartz-like consistency. A mechanical rate depends on temperature, position, winding state, magnetism and age, and it will move by a second or two a day for reasons you cannot control. A watch that averages plus 3 seconds a day for a month and then plus 5 is not developing a fault. It is a machine with a metal spring in it, behaving like one.
Common follow-up questions
How long should I test a watch before deciding it is inaccurate?
Seven days of normal wear, minimum. A shorter test mixes your own reading error and normal day-to-day variation into the result in a way you cannot separate. If the daily figures vary widely, run a second week and note your activity and overnight resting position each day, because the explanation is usually there.
Do I need to buy a timegrapher?
No. A timegrapher answers questions about the movement's health, principally amplitude, that the wrist test cannot, but it does not tell you how the watch will perform on your wrist. If you own several mechanical watches or buy pre-owned regularly, an inexpensive machine is genuinely useful for spotting a movement that needs a service. For a single watch, the seven-day test plus a watchmaker's opinion covers it.
Why does my watch show a different rate on a timegrapher than on my wrist?
Because they measure different things. The machine reports an instantaneous rate with the watch clamped in one position at one state of wind. Your wrist presents a changing sequence of positions, temperatures and winding states, and the result is an average of all of them. A watch reading plus 1 dial up on a machine can easily average plus 6 in real wear.
Is a high beat error actually a problem?
Below about 1.0 ms, usually not for timekeeping. Its practical effect shows at low amplitude, where a badly out-of-beat watch can fail to restart after a knock or after running down. Below about 0.5 ms is considered good, and a figure much above 1.0 ms is worth correcting at the next service rather than as a separate job.
Can regulation fix a watch that keeps different time every day?
No. Regulation moves the whole rate up or down; it cannot make an inconsistent watch consistent. Day-to-day inconsistency comes from positional variation, irregular winding, magnetism or a mechanical problem, and each has a different fix. Measure first: paying to have a watch regulated when the real issue is low amplitude simply moves the problem.
Sources and further reading
- ISO 3159, Timekeeping instruments: wrist-chronometers with spring balance oscillator, for the multi-position, multi-day test protocol and the minus 4 to plus 6 seconds per day tolerance.
- Contrôle Officiel Suisse des Chronomètres (COSC), published description of its testing criteria including mean daily rate and difference between positions.
- Donald de Carle, Practical Watch Repairing, for regulation, beat setting and the function of the regulator arm, curb pins and hairspring collet.
- George Daniels, Watchmaking (Philip Wilson Publishers), for escapement behavior, amplitude and the relationship between torque and rate.
- Movement manufacturers' technical data sheets (ETA, Sellita, Seiko, Citizen), which publish the lift angle, nominal frequency and rate tolerances used when setting up a timing machine.
Last reviewed 4 September 2026. Spotted an error? Tell us and we will fix it in public.