Key takeaways
- The accepted value is the denominator. Dividing by the measured value flips the comparison and inflates the result when the measurement is small.
- Percent error is unsigned by convention. The bars around the difference mean an error of −2% and +2% both report as 2%.
- Percent difference is for two measurements. With no accepted value, compare to the average of the pair instead.
- Percent error is not percent change. Change uses the old value as the base and keeps its sign.
- Small accepted values amplify the number. The same absolute miss can be negligible against 5,000 and huge against 0.05.
What percent error compares
Percent error is a unitless way to describe how far a measurement sits from a reference value. The reference is usually a textbook constant, a certified standard, or a value from a more precise instrument. Because the result is a percentage, a 0.1 cm miss on a 10 cm object and a 0.1 cm miss on a 1,000 cm object can be compared fairly.
It is a diagnostic, not a complete uncertainty statement. Percent error measures the distance from one accepted number, while measurement uncertainty describes a range of plausible values. A result with 2% error is reassuring only if the instrument and the accepted value are both better than 2% to begin with.
The formula and each part
Percent error = |measured − accepted| ÷ |accepted| × 100
The numerator is the absolute size of the discrepancy; the denominator scales it to the accepted value; the 100 converts the fraction to a percentage. Some textbooks write (measured − accepted) ÷ accepted × 100 without bars and report a signed result, where a negative answer means the measurement came in below the accepted value. Both conventions are common, so state which one you used.
The denominator is where most errors happen. Percent error is always relative to the accepted value, because that is the number being treated as the truth. Dividing by the measured value instead changes the answer, sometimes by several percentage points when the two values differ substantially.
Worked example: measuring density
A metal sample has an accepted density of 2.70 g/cm³. A student measures 2.65 g/cm³. The difference is |2.65 − 2.70| = 0.05 g/cm³, so the percent error is 0.05 ÷ 2.70 × 100 = 1.85%.
The sign tells a story the absolute value hides: the measured density is low, which is consistent with a volume reading that was slightly too large or a mass reading that was slightly too small. Repeating the measurement three times and averaging would reduce random scatter, but it would not remove a systematic bias such as a miscalibrated balance. That is why a percent error close to zero is not proof of a good method — it can also come from two errors that cancel.
Worked example: checking a scale
A 50.0 g calibration weight reads 50.7 g on a bench scale. Percent error = |50.7 − 50.0| ÷ 50.0 × 100 = 0.7 ÷ 50.0 × 100 = 1.40%. The scale reads high by 1.4%, so every subsequent mass on that scale inherits the same proportional bias.
If the scale’s specification is ±0.5%, a 1.4% error is nearly three times the allowed tolerance and the instrument needs adjustment. If the specification is ±2%, the same reading is acceptable. Percent error means little without the tolerance it is being compared against, which is why calibration certificates pair the two numbers.
Percent error versus percent difference
Percent difference compares two measured values when neither is authoritative: |a − b| ÷ ((a + b) ÷ 2) × 100. Two trials reporting 9.8 and 10.2 give |9.8 − 10.2| = 0.4 and an average of 10.0, so the percent difference is 4.0%.
Percent error answers “how far from the truth?”; percent difference answers “how far apart are these two?” Using percent error with one of the measurements as the denominator quietly treats that measurement as correct, which is exactly the assumption percent difference is designed to avoid. In a lab report, choose the formula based on whether a reference value exists.
Percent error versus percent change
Percent change tracks movement over time: (new − old) ÷ old × 100, and the sign matters. A price rising from 80 to 92 is a 15% increase. Percent error would compare 92 with an accepted 80 and report 15% as well, but the meaning is different — one is growth, the other is a discrepancy.
Percent change can also be larger than 100% or negative without any implication of a mistake, while percent error above 100% usually signals a serious problem: a measurement of 250 against an accepted 100 is a 150% error, meaning the reading is two and a half times the reference. Keeping the two formulas separate prevents a normal month-over-month change from being reported as an experimental failure.
Accuracy, precision, and uncertainty
Accuracy is closeness to the accepted value, which is what percent error measures. Precision is closeness of repeated measurements to each other, which percent error says nothing about. A tight cluster of readings 5% high is precise and inaccurate; a wide scatter centered on the true value is accurate on average and imprecise.
Uncertainty describes the range around a result. If a length is reported as 12.0 ± 0.1 cm, the ±0.1 cm is the uncertainty and the 12.0 cm is the estimate. Percent error against a reference of 12.0 cm is 0%, yet the measurement is still only good to about 1 part in 120. NIST guidance treats the uncertainty statement as the primary result and the error as a comparison, not a substitute.
What counts as a good percent error
There is no universal threshold. Analytical chemistry methods are often expected within 1–2% of a certified reference material; undergraduate labs commonly accept 5%; engineering estimates and field measurements may tolerate 10% or more. The right question is whether the error is smaller than the tolerance the decision requires.
Compare the error against the instrument’s specification and the accepted value’s own uncertainty. If a balance is rated ±0.1 g and you weigh 5.0 g, the instrument alone can produce a 2% error, so demanding 0.5% agreement is unrealistic. A percent error smaller than the combined uncertainty of the inputs is not evidence of exceptional skill; it is within the noise.
Common mistakes
- Dividing by the measured value instead of the accepted value.
- Forgetting to multiply by 100, then reporting 0.0185 as “1.85%.”
- Taking the absolute value when direction matters, or keeping a sign when the convention calls for an absolute error.
- Comparing quantities in different units, such as grams against kilograms, without converting first.
- Using percent error when there is no accepted value; percent difference is the correct tool.
- Reporting six decimal places from inputs measured to two significant figures.
- Assuming a low percent error proves the method is sound, when two biases may be cancelling.
Practical caveats
Small denominators distort the number. An absolute miss of 0.002 against an accepted 0.050 is a 4.0% error, while the same 0.002 against 5,000 is 0.00004%. Always report the absolute difference alongside the percentage so the reader can judge the scale.
The accepted value is not infinitely precise. A constant quoted to three significant figures carries its own uncertainty, and comparing a four-digit measurement against a three-digit reference caps how meaningful the last digit can be. Round the percent error to the precision the inputs support — usually one or two significant figures — and state the formula and the reference value you used.
Frequently asked questions
Can percent error be negative?
With the absolute-value formula, no: it is zero or positive. If you use the signed version, a negative result means the measurement is below the accepted value, not that the error is smaller.
What is the difference between percent error and percent difference?
Percent error compares one measurement with an accepted value. Percent difference compares two measurements with no established truth, using their average as the denominator.
Can percent error be greater than 100%?
Yes. A measured 250 against an accepted 100 gives 150%. Values above 100% simply mean the measurement is more than twice the reference, and they usually indicate a units or setup problem.
Should I use the absolute value or keep the sign?
Use the absolute value when you only care about the size of the discrepancy. Keep the sign when the direction matters, such as a scale that reads consistently high or a model that underpredicts.
Is a lower percent error always better?
No. A low error can come from luck or from two systematic errors cancelling. Precision, repeatability, and an honest uncertainty statement matter more than a single impressive percentage.
How many decimal places should I report?
Match the inputs. If the accepted value has three significant figures and the measurement has three, one decimal place in the percentage is usually enough; reporting more implies precision the data does not have.
Further reading