Percent Difference Calculator – How Far Apart Two Values Really Are

Compare up to four values at once

Got three or four quotes, readings or scores? Enter them all and see every pairwise percent difference in one symmetric matrix — plus which pair sits closest together and which is spread furthest apart.

Two values, one fair midpoint

Value 1 in violet, Value 2 in peach — and the dashed line marks their average, the base this calculator divides by.

Keep exploring

Three companion tools for percentages.

How it works

Three steps from two values to one fair comparison.

The math behind percent difference

Go deeper

When the average is the right base — and when it is not.

Percent difference, percent change, percent error and percent increase — four tools, four questions

One gap, four divisors, four different answers. The values 80 and 120 sit 40 apart, and every tool below starts from that same gap of 40. What changes is the base it is divided by, and the base is the whole story: pick the wrong base and a correct calculation still answers the wrong question. The table below puts the four tools side by side on the same pair, so the difference between them is visible instead of theoretical.

Percent increase deserves a careful note, because it is often treated as a separate tool when it is really percentage change with a restricted sign. An increase from 80 to 120 is a percentage change of +50.00%, and calling it a 50.00% increase says the same thing in plainer words. The reverse move, 120 down to 80, is a percentage change of −33.33%, and it is described as a 33.33% decrease, not a negative increase. The asymmetry is the point: a 50% rise followed by a 50% fall does not return to the start, because the second 50% is taken from the larger, risen value. From 80, a 50% rise reaches 120, and a 50% fall from 120 lands at 60, not 80. Percent difference never has this problem, because it has no direction to reverse: 80 and 120 are 40.00% apart whichever way the pair is read.

Percent error is the tool for the case where one value has authority. A laboratory instrument reads 120 for a standard known to be 100, and the error is the gap of 20 divided by the true value of 100, giving 20.00%. The percent difference between the same two numbers is 20 divided by their average of 110, giving 18.18%, and that number answers a different question: how far apart are two readings of equal standing. When a true value exists, error is the honest tool, because it measures against the standard rather than splitting the difference with it. When neither value is the standard, as with two competing lab methods or two sellers quoting different prices, difference is the honest tool, because dividing by either value alone would quietly promote it to the role of standard.

The decision checklist is short. Are the two values peers, with neither one first and neither one true? Use percent difference. Did one value genuinely come before the other, as in last year versus this year, or a price before and after a rise? Use percentage change, and name the direction. Is one value the accepted, true or target value? Use percent error. Is the move upward and the audience expects growth language? Percent increase, which is percentage change for rises, stated as a rise. The Advanced options panel on this page keeps all four visible at once for the pair above: the signed difference, the step working, both directions of percentage change, and the percent error with Value 1 as the true value, so a reader can watch the same gap produce four numbers and learn which question each one answers.

A final confusion to retire is the idea that percent difference and percentage change are two names for one thing, or that one of them is the corrected version of the other. They are not rivals. A salary that rises from $80,000 to $120,000 grew by 50.00%, and that is the right number for the negotiation, the budget and the headline. The same two salaries, compared as two offers on the table with no timeline, are 40.00% apart, and that is the right number for choosing between them. The calculation is trivial in both cases. The judgment is choosing the base, and the base follows the question, never the other way round.

Step-by-step working: the average denominator, signed values and negative numbers

The formula looks compact, and compact formulas hide their decisions. Worked slowly, the percent difference between 80 and 120 runs in three steps, and the Advanced options panel on this page shows the same three steps live for whatever pair is entered. Step one is the gap: subtract one value from the other and take the size of the result, so |120 − 80| = 40. The absolute value matters, because the gap is a distance, and distances are never negative. Step two is the average of the sizes: (|120| + |80|) ÷ 2 = 100. Note that the average is taken over the sizes, the absolute values, not over the signed values, which is the detail that makes the formula behave with negative numbers, as shown below. Step three is the division and the scaling: 40 ÷ 100 × 100 = 40.00%. Nothing else happens. Every percent difference on this page is those three steps, and every worked example below can be checked by repeating them.

The signed difference is the same calculation with the absolute value removed from the gap only. Signed percent difference equals (Value 2 − Value 1) divided by the average, times 100. For 80 and then 120, that is (120 − 80) ÷ 100 × 100 = +40.00%. Reverse the order and it is (80 − 120) ÷ 100 × 100 = −40.00%. The magnitude is identical to the absolute answer, and the sign now carries the direction: positive means Value 2 sits above Value 1, negative means it sits below. The signed form is useful when the order of the pair still means something, for example two readings taken in sequence where the direction of drift matters, but neither reading is a starting point in the percentage-change sense. The toggle in the Advanced options panel switches the headline readout between the absolute and signed forms without touching the main result, so both views stay available.

Negative numbers are where the average-of-sizes detail earns its keep. Take −40 and −60. The signed average of those two values is −50, and dividing a gap of 20 by −50 would give a nonsense negative percentage. The formula instead averages the sizes, 40 and 60, giving 50, and the percent difference is 20 ÷ 50 × 100 = 40.00%, a sensible answer: two negative values 20 apart around a typical size of 50. The same logic handles pairs that straddle zero. Take −20 and 30. The gap is |30 − (−20)| = 50, the average of the sizes is (20 + 30) ÷ 2 = 25, and the percent difference is 50 ÷ 25 × 100 = 200.00%, the maximum of the scale. That maximum is not a coincidence: when the values sit on opposite sides of zero, the gap equals the sum of the sizes, and the sum of the sizes is exactly twice their average, so the ratio is exactly 2, or 200%. A temperature pair such as −10 and 10 behaves the same way: gap 20, average size 10, difference 200.00%. The result looks extreme, and it is honest, because a pair straddling zero really is as far apart, relative to its size, as any pair can be.

Zero itself has two cases, and they must not be merged. One value zero and the other not zero gives 200.00%, as 0 and 100 show: gap 100, average 50, ratio 2. Both values zero is undefined, because the average is 0 and division by zero has no answer; the calculator says so plainly instead of printing a number. That matches the disclaimer on this page, and it is the only input pair for which the tool refuses an answer. Blank fields are not an error in the same sense: the pair calculator asks for both values, while the four-value comparison simply leaves blank entries out of its matrix.

Rounding is the last working detail, and the rule is to round once, at the end. The pair 100 and 130 has a gap of 30 and an average of 115, and 30 ÷ 115 = 0.260869…, which rounds to 26.09%. Rounding the gap or the average first would shift the answer in the second decimal place, which is why the calculator keeps full precision internally and formats to two decimals only for display. A reader checking by hand should do the same: carry the division out to several places, then round the percentage. When a hand check disagrees with the calculator by 0.01%, the cause is almost always an intermediate rounding, not a different formula.

Comparing three or four values: reading the matrix and ranking every pair

Two values give one percent difference. Four values give six, because every pair of values has its own difference, and the comparison matrix on this page shows all of them at once. The matrix is symmetric: the entry for A against B is the same as B against A, which is the whole point of percent difference, and the diagonal is a dash, because a value compared with itself is 0% apart and the dash keeps the table readable. The closest pair and the furthest pair are picked out in the chips below the matrix, and the ranking panel underneath lists every pair from closest to furthest, so the structure of a set of values is visible in one glance rather than six separate calculations.

A worked four-value set makes the reading concrete. Take the values 50, 90, 130 and 200. The six pairs, each worked by the same three steps as the pair calculator, are: 50 and 90 differ by 40 over an average of 70, giving 57.14%; 50 and 130 differ by 80 over an average of 90, giving 88.89%; 50 and 200 differ by 150 over an average of 125, giving 120.00%; 90 and 130 differ by 40 over an average of 110, giving 36.36%; 90 and 200 differ by 110 over an average of 145, giving 75.86%; and 130 and 200 differ by 70 over an average of 165, giving 42.42%. The table below repeats those figures so they can be checked against the live matrix.

The ranking tells a story the raw values hide. The closest pair is 90 and 130, at 36.36% apart, and the furthest is 50 and 200, at 120.00% apart, which is no surprise, since they are the smallest and largest values. The interesting entry is the middle: 50 and 90 are 57.14% apart even though their gap is only 40, the same gap as 90 and 130, because their average is much smaller, 70 against 110. Equal gaps are not equal differences, and that is exactly why the matrix divides by the average instead of reporting raw gaps. A reader who only looked at gaps would call 50-to-90 and 90-to-130 equally close. They are not: relative to their size, the lower pair is far more spread out.

Practical use follows the same reading. Four supplier quotes for the same item, four lab readings of one sample, or four test scores for one student all become one matrix, and the ranking answers the first question about any such set: do these values agree, or is one of them an outlier? A set where every pair sits under 10% is a set in agreement. A set where one value is 50% or more from all the others has an outlier, and the ranking shows which value it is, because the outlier owns the largest entries. The default values on this page, 80, 100 and 120, show the agreement case in miniature: the pairs are 22.22%, 18.18% and 40.00%, with 100 and 120 closest together, so the set is spread but not wild.

Two cautions belong with the matrix. First, the closest pair is not the best pair. Two expensive quotes can be close together while both are far above a cheaper third quote, and closeness measures agreement, not value. The ranking says which values cluster, and the reader still has to judge whether the cluster is where they want to be. Second, adding a fourth value adds three new pairs, not one, so a four-value matrix deserves a slower read than a three-value one: six comparisons, each with its own base, and the largest entry is the one that usually matters most. Blank entries are simply left out, so a comparison can start with two values and grow as more arrive, and the matrix rebuilds itself each time without any button press.

Where percent difference belongs: lab results, prices, scores and the mistakes to avoid

Percent difference earns its place wherever two numbers of equal standing need one fair comparison. The laboratory is the classic case. Two methods measure the same sample and return 98.4 and 101.2. The gap is 2.8, the average is 99.8, and the percent difference is 2.8 ÷ 99.8 × 100 = 2.81%. Neither method is the true value, so percent error has nothing to divide by, and percentage change would arbitrarily crown the first method as the starting point. The 2.81% figure is the number a lab notebook wants: small enough to call the methods consistent, large enough to record. The same reasoning covers repeated measurements on one instrument, where the question is drift and agreement rather than accuracy against a standard.

Prices are the everyday case, and the page’s own example carries it: two sellers quote $80 and $120 for the same item, and the quotes are 40.00% apart. A shopper comparing three or four quotes uses the matrix instead, and the ranking shows whether the market is tight or scattered. Timing in sport works the same way. Two sprint times, 10.2 seconds and 9.9 seconds, have a gap of 0.3 over an average of 10.05, giving 2.99%. Test scores follow too: 72 and 84 differ by 12 over an average of 78, giving 15.38%. In each case the values are peers, and the average base stops either side from framing the comparison in its own favour. A seller quoting the $80 price would rather say the rival is 50% more expensive, dividing by 80, while the $120 seller would rather say the rival is 33.33% cheaper, dividing by 120. Percent difference refuses both framings and gives 40.00%, which is why it is the number both sides can accept.

Scale intuition helps when a result needs a sanity check without a calculator. A percent difference of 10% means the larger value is about 1.11 times the smaller one, as 90 and 100 show: 10.53% for a ratio of 1.11. A difference of 40% means a ratio of 1.50, the 80 and 120 case. A difference of 100% means a ratio of exactly 3, so any pair further apart than 3-to-1 passes 100%, and 10 against 50, a ratio of 5, reaches 133.33%. At the far end, 200% is the ceiling, reached only when one value is 0 or when the pair straddles zero. These landmarks let a reader glance at a result and know whether it is plausible: a claimed 60% difference for values in a 1.2 ratio is wrong, because a 1.2 ratio gives 2 × 0.2 ÷ 2.2, which is 18.18%, not 60%.

The mistakes that matter most are all base mistakes, not arithmetic mistakes. The first is quoting percentage change as if it were percent difference, which overstates every comparison where the smaller value is the starting point: 80 to 120 is a 50% change, not a 50% difference. The second is averaging percent differences. Two pairs at 40% and 20% do not average to a 30% difference for the combined set, because each percentage has a different base; the correct move is to return to the values and compute the new pair directly. The third is using difference where a true value exists, which understates error by dividing by the average instead of the standard: a reading of 120 against a true 100 is 20.00% in error, not 18.18% different. The fourth is dividing by one value out of habit, which is just percentage change wearing the wrong label. And the fifth is forgetting that both-zero is undefined rather than 0%: identical non-zero values are 0% apart, but 0 and 0 have no average to divide by, and the honest answer is that the comparison does not exist.

None of this makes percent difference a universal tool. It is deliberately narrow: it compares exactly two values, or a small set of them, as equals, with no timeline and no standard. Where a timeline exists, percentage change is better. Where a standard exists, percent error is better. Where a whole distribution needs summarising, a mean and a spread are better. The strength of percent difference is that, inside its narrow job, it is fair in a way the alternatives are not: one number, the same in both orders, owned by neither side. That fairness is the reason the formula divides by the average, and the reason this page keeps the average visible in every result, every step and every worked example.

Frequently asked questions

Quick answers about percent difference.

Related calculators

More tools for percents, averages and ratios.