Percent Error Explained: Formula, Examples and Common Mistakes

What percent error actually measures

Percent error is a single number that says how close a measurement got to a value you already accept as true. It is the standard accuracy metric in physics labs, chemistry coursework, calibration reports, and engineering tolerance checks — any situation where you can compare a reading against a reference and want to express the gap as a percentage rather than in raw units. The percent error calculator at the top of this page takes the measured and actual values and returns the percent error along with the signed version, which tells you whether the measurement came in high or low.

The word "error" in the name is a frequent source of confusion. It does not imply you did something wrong — every real measurement has some error, because every instrument has finite resolution and every observer has finite reaction time. Percent error simply quantifies the gap between observation and accepted value. A lab thermometer that reads 99.2 °C for boiling water at sea level has a percent error of 0.8%, which is a perfectly reasonable result for a glass thermometer with 0.5 °C graduations. The same percent error from a calibrated platinum resistance thermometer would suggest the instrument is broken.

The formula and where each piece comes from

The textbook definition is:

percent error = |measured − actual| / |actual| × 100

The numerator is the absolute difference between what you measured and the value you are comparing to. Taking the absolute value converts the difference into a positive magnitude, so percent error is reported as a non-negative number — a measurement that is 5% too high and one that is 5% too low both yield a percent error of 5%. If you want to preserve direction, you keep the sign on the numerator; the result is sometimes called signed percent error or percent deviation, and the percent error calculator shows both forms.

The denominator is the absolute value of the actual figure. The absolute value matters when the reference is negative — a temperature of −10 °C, an electric potential of −2.3 V, a debit balance of −£500. Without the absolute value in the denominator, a perfectly accurate measurement of a negative quantity would produce a negative percent error, which is nonsense. NIST Technical Note 1297, the long-standing reference on expressing uncertainty in physical measurements, formalises this convention; Taylor's An Introduction to Error Analysis(1997) gives the same definition in §1.6.

Multiplying by 100 just converts the fraction into a percentage. Some sources skip the multiplication and report a fractional error on the 0–1 scale instead; the underlying number is the same, only the units differ. A fractional error of 0.008 and a percent error of 0.8% describe an identical measurement.

Worked example: a chemistry lab density measurement

A first-year chemistry student measures the density of a copper sample by weighing it and using water displacement to find its volume. The mass comes out to 89.3 g and the volume to 10.2 mL, so the calculated density is 89.3 / 10.2 = 8.755 g/mL. The accepted density of pure copper at 20 °C is 8.96 g/mL (CRC Handbook of Chemistry and Physics, 102nd edition). What is the percent error?

Step by step:

1. Absolute error: |8.755 − 8.96| = 0.205 g/mL.
2. Divide by the absolute actual value: 0.205 / 8.96 = 0.02288.
3. Multiply by 100: 2.29% (rounded to two decimals).
4. Signed percent error: (8.755 − 8.96) / 8.96 × 100 = −2.29%, telling you the measurement underestimated the true density.

Feed those numbers into the percent error calculator directly to see the same result, plus the absolute error and signed error broken out explicitly. For a high-school or introductory undergraduate lab a 2.3% error on a density measurement is well within the expected range — most of the error in this kind of experiment comes from reading a graduated cylinder to ±0.1 mL, which alone accounts for about 1% uncertainty on a 10 mL volume.

Factors that drive percent error up

Instrument resolution

Every instrument has a smallest division. A ruler graduated in millimetres cannot resolve sub-millimetre differences, a digital scale that displays whole grams cannot tell 5.4 g from 5.6 g, and an oscilloscope sampling at 1 ns has a floor on how precisely it can time-stamp an event. The minimum percent error you can hope to achieve is bounded below by instrument resolution divided by the measured value. A 1 mm ruler used to measure a 10 mm object will produce errors on the order of 10%; the same ruler used to measure a 1000 mm object lives at around 0.1%.

Systematic offset (calibration drift)

A miscalibrated instrument adds a fixed bias to every reading. If your kitchen scale is consistently 3 g heavy, every measurement carries that offset — and the resulting percent error gets larger as the measured quantity gets smaller. A 3 g bias on a 300 g flour weigh-out is 1%; on a 30 g spice weigh-out it is 10%. Calibrating the instrument against a known standard removes the offset and is the single highest-leverage intervention available to most experimenters.

Random noise

Random fluctuations are the part of measurement error that does not have a consistent direction. Air currents move a balance pan up and down by a few milligrams, thermal noise in an analogue voltmeter jitters the reading, vibrations from a passing lorry shift a microscope's focus. Random noise is what repeated measurements average out — taking ten readings and using the mean typically reduces the noise contribution to percent error by a factor of about √10, or roughly 3×. The standard deviation calculator gives you the spread of repeated readings, which is the right number to use when deciding how many trials are enough.

Procedural error

Procedural errors come from how a measurement is performed rather than the instrument itself. Parallax when reading a meniscus, starting a stopwatch a fraction of a second late, failing to zero a balance before weighing, taking a reading before a system has reached equilibrium — all of these add a per-trial error that does not show up in the instrument's spec sheet. Procedural errors are usually the dominant contribution to percent error in an introductory teaching lab, which is why labs spend so much time on technique.

Reference uncertainty

The "true" value you compare to has its own uncertainty. The molar mass of carbon-12 is exactly 12 g/mol by definition, but the molar mass of natural copper is 63.546 ± 0.003 g/mol — the last digit reflects isotopic abundance variation in different ore bodies. Comparing a high-precision measurement against a reference whose published uncertainty is similar in size means the calculated percent error mostly measures the noise in the reference, not your experiment. For coursework this rarely matters; for metrology and standards work it is central.

How to reduce percent error

• Calibrate before each session. A two-point calibration against known standards (ice water at 0 °C and boiling water at 100 °C for a thermometer; tared empty flask and 100 g reference mass for a balance) removes the systematic offset that dominates most measurements.
• Take repeated readings and average them. The mean of n independent trials reduces the random component of error by √n. Three trials cut it by 1.7×; ten by 3.2×; a hundred by 10×. Diminishing returns kick in quickly — past ten trials you are usually limited by systematic, not random, error.
• Use an instrument matched to the scale you are measuring. A micrometer beats a ruler for sub-millimetre lengths; an analytical balance beats a kitchen scale for sub-gram masses. Resolution should be at least 10× finer than the precision you need in the final result.
• Control the environment. Draughts, temperature swings, vibrations, and electromagnetic interference all add noise. A balance in a glass enclosure, a voltmeter on a shielded bench, and a thermometer given time to equilibrate before reading all reduce random error meaningfully.
• Reduce the effect of parallax and reaction time. Read scales at eye level, use a mirror behind a needle gauge, and time experiments with an electronic trigger rather than a stopwatch where possible. These changes feel fussy but reliably trim half a percent from typical undergraduate measurements.
• Compare to the best available reference. If your textbook value disagrees with a more recent CODATA or NIST value, use the newer figure. Percent error is only as good as the "truth" in the denominator.

Common mistakes

Confusing percent error with percent difference. Percent error compares a measurement to an accepted truth. Percent difference compares two measurements with no privileged reference and uses their average in the denominator: |a − b| / ((a + b) / 2) × 100. If you compute your two trials with the percent-error formula you will pick one of them as the "true" value, which is arbitrary and gives different answers depending on which one you choose.

Forgetting the absolute value in the denominator. When the actual value is negative, omitting the absolute value flips the sign of the result and produces a negative percent error, which is meaningless. The magnitudes-based definition handles negative references correctly; the signed version preserves direction in the numerator but still uses the absolute value in the denominator.

Reporting more significant figures than the inputs justify.Percent error inherits the precision of its inputs. If the measured value has three significant figures and the actual value has four, the percent error has at most three. Reporting "2.293587%" from inputs of 8.76 and 8.96 suggests a precision your measurement does not have — round to 2.3% or 2%.

Treating percent error as a measure of precision. Percent error is an accuracy metric — it tells you how close one measurement was to the true value. Precision is about how repeatable measurements are, regardless of whether the cluster of readings is centred on the truth. A target with all the holes clustered tightly an inch above the bullseye is precise but inaccurate; one with the holes scattered around the bullseye is accurate on average but imprecise. To describe both you need percent error plus a spread statistic such as standard deviation.

When to seek a more rigorous treatment

Percent error is the right tool for a single measurement compared against a known value. For more demanding work — publishable physics results, calibration certificates, regulatory submissions — you usually need to propagate uncertainties through the calculation, distinguish Type A (statistical) from Type B (systematic) uncertainty as in GUM (the Guide to the Expression of Uncertainty in Measurement), and report results as value ± expanded uncertainty with a coverage factor. Percent error remains useful as a quick sanity check even in that context, but it stops being sufficient on its own. Lab reports in introductory courses, quality-control checks against a working standard, and rough back-of-envelope assessments are where percent error is the right level of rigour.

Frequently asked questions

What counts as a "good" percent error?

It depends entirely on context. For a high-school chemistry lab, anything under 5% is usually considered acceptable. For introductory undergraduate physics, the target is often under 2% for length and mass measurements and under 5% for timing and electrical measurements. In metrology — the science of measurement itself — percent errors are often expressed in parts per million, and 0.001% is a routine target for primary standards. The right benchmark is the published uncertainty of the instrument and the historical reproducibility of the measurement, not a fixed numerical threshold.

Can percent error be larger than 100%?

Yes — if the measured value differs from the actual value by more than the magnitude of the actual value. Measuring 12 against a true value of 5 gives a percent error of 140%. Errors above 100% almost always indicate something bigger than ordinary noise: a unit-conversion mistake (grams reported as kilograms), a missing decimal point, a transposed digit, or a confused setup (measuring the wrong sample). When percent error is unexpectedly large, the first instinct should be to look for these gross errors before assuming the instrument is at fault.

What does it mean if my percent error is exactly zero?

Either the measurement happened to fall on the accepted value to the precision you wrote it down, or — more commonly in coursework — the inputs were rounded so aggressively that genuine error was hidden by the rounding. A density measurement reported as 9 g/mL against an accepted value of 9 g/mL gives a percent error of 0%, but if the underlying readings were 8.7 g/mL and 9.3 g/mL, the true percent error is 3% or so. Carry one extra significant figure through the calculation and round only at the end.

How is percent error different from absolute error?

Absolute error is the raw size of the gap, expressed in the same units as the measurement: |measured − actual|. Percent error scales that gap by the actual value and expresses it as a percentage. A 1 g absolute error on a 10 g sample is a 10% percent error; the same 1 g error on a 10 kg sample is 0.01%. Absolute error tells you how far off the measurement was; percent error tells you how large the gap is relative to the thing being measured, which is usually what you care about when comparing measurements across scales.

Why does my answer differ from the textbook's by a factor of 10 or 100?

Almost always a units problem. Two common cases: mixing percent and decimal (treating 5 and 5% as interchangeable), or comparing values in different units (g/mL against kg/m³, mph against m/s). The fix is to write the units explicitly at every step of the calculation and check that they cancel correctly. The percent error calculator assumes both inputs share the same units; if they do not, the percentage it returns will be off by exactly the unit-conversion factor.

Should I always report the signed percent error or the magnitude?

The convention in physics and chemistry coursework is to report the magnitude (the absolute-value version) by default, because percent error is treated as a size statistic. The signed version is useful when direction matters — for example, when you want to detect a consistent bias in a series of measurements, or when you are comparing two different methods to see which systematically overshoots. The calculator displays both so you can pick whichever suits the report.

What about percent error when the actual value is zero?

Percent error is undefined when the reference value is zero, because the formula divides by zero. The only meaningful quantity in that case is the absolute error, |measured − 0| = |measured|. If both the measured and actual values are zero, percent error is conventionally taken to be 0%. The calculator detects the zero-reference case and shows the absolute error rather than producing an infinite or meaningless percentage.

Related calculators

• Standard deviation calculator — the spread of repeated measurements; pairs with percent error to describe both accuracy and precision.
• Average calculator — mean, median and mode of repeated trials, the input you need before computing percent error from averaged data.
• Confidence interval calculator — the margin of error around a sample mean, useful when you want a range rather than a single percent-error figure.
• Sample size calculator — how many trials are needed to bring random error below a chosen threshold.