Every number carries more precision than most situations actually need — a bank balance might sit at $412.386219 in raw floating-point arithmetic, but nobody's till drawer deals in fractions of a tenth of a cent. Rounding is the deliberate act of throwing away precision you don't need, in exchange for a simpler number that's close enough to trust — "close enough" being defined by whichever decimal place or significant figure you've decided is the cutoff.
Why the Halfway Point Is the Natural Cutoff
Whatever digit sits just past your rounding cutoff tells you which side of the halfway mark the true value falls on. A digit of 5 or higher means you've passed the midpoint toward the next value up, so rounding up gets you closer to the truth; a digit below 5 means you're still closer to where you started, so staying put is the more accurate choice. That's the entire logic — round toward whichever nearby value you're actually closest to.
Worked example: round 6.283 to two decimal places. Past the second decimal spot sits a 3 — under the halfway mark — so nothing changes: 6.28.
Worked example: round 6.287 to two decimal places. This time a 7 sits just past the cutoff, over the midpoint, so the target digit bumps up: 6.29.
The same halfway-mark logic scales up to whole-number place values without any modification. Rounding 47,382 to the nearest thousand looks at what's sitting just past the thousands digit — a 3, under the midpoint — so it rounds down to 47,000. Rounding 583 to the nearest ten checks the digit past the tens place — another 3 — rounding down to 580.
Significant figures shift where you start counting from, not the underlying rule. Instead of anchoring to the decimal point, you anchor to the first digit that isn't a zero. Rounding 0.03847 to 2 significant figures means the two digits that count are 3 and 8; what follows (4) sits under the midpoint, so it stays 0.038. Rounding 5,672 to 3 significant figures keeps 5, 6, 7 as the meaningful digits; the next one (2) is under the midpoint too, rounding down to 5,670.
Round Half Up vs. Round Half to Even
Nothing in the halfway-mark logic above resolves what happens when the leftover digit is exactly 5 with nothing after it — "5 rounds up" is simply the convention this guide and most classrooms settle on, not a mathematical necessity. Statisticians and bankers sometimes prefer "round half to even" instead, nudging a tied value toward whichever neighboring digit happens to be even — 2.5 becomes 2, while 3.5 still becomes 4 — because over thousands of repeated roundings, always breaking ties upward quietly inflates the total, and always rounding toward even cancels that drift out. Neither convention is "more correct"; they simply diverge on this one exact tie.
Rounding versus truncating — two different operations, easy to conflate. Truncating simply discards digits past a certain point without applying the up/down decision at all. Truncating 8.79 to one decimal place gives 8.7 (discarding the 9 regardless of its value), while properly rounding 8.79 to one decimal place gives 8.8. Some display contexts (like a countdown timer showing whole seconds) intentionally truncate; most everyday numeric reporting expects true rounding.
Rounding in one step versus rounding in stages. Round directly from the original number to your final target precision whenever possible. Rounding in successive stages can produce a different, wrong answer: rounding 2.449 to one decimal place first gives 2.4 (since the second decimal digit, 4, is below 5), and rounding 2.4 to a whole number then gives 2. But rounding 2.449 directly to a whole number also gives 2 in this case — the two methods happen to agree here, though they don't always; a value like 1.45 rounded to one decimal place first (1.5, since 5 rounds up) and then to a whole number (2) gives a different answer than rounding 1.45 directly to a whole number (1, since the digit after the ones place is 4, below 5). The two-stage version is a genuine trap.
Common Mistakes When Rounding
Rounding in stages instead of directly, as shown above, is the most subtle and common error. A second common mistake is applying decimal-place rounding logic when significant-figure rounding was actually called for (or vice versa) — the two use a different starting reference point (the decimal point versus the first nonzero digit) and give different results for numbers with leading zeros, like 0.0048.
Worked example: round 12,499 to the nearest hundred. The digit right after the hundreds place is 9 (12,499), which is ≥5, so round the hundreds digit up: from 4 to 5, giving 12,500. Note this changes multiple digits at once due to carrying — the hundreds digit rolling from 4 to 5 doesn't affect anything else here, but rounding a number like 995 to the nearest ten does cascade (the digit after the tens place is 5, so round the tens digit up from 9 to 10, which carries into the hundreds place, giving 1,000, not 9(10)).
Worked example: round 3.14159 (an approximation of pi) to 4 significant figures. The first four significant digits are 3, 1, 4, 1; the next digit (5) is ≥5, so round the fourth digit up: 3.142.
A practical everyday scenario where the rounding convention genuinely matters: splitting a bill. If $50.03 is split three ways, each person's exact share is $16.6766..., and every reasonable rounding approach (round to the nearest cent) gives $16.68 per person — but three people paying $16.68 each totals $50.04, one cent more than the actual bill. This kind of small rounding discrepancy is common and usually resolved by having one person absorb the extra cent, a reminder that rounded parts don't always sum back exactly to a rounded (or even an unrounded) whole.
Worked example: round a large currency figure for a headline-style summary, 4,832,910 to the nearest million. The digit right after the millions place is 8 (4,832,910), which is ≥5, so round the millions digit up: from 4 to 5, giving 5,000,000. This kind of large-scale rounding is exactly what news headlines do when reporting "roughly $5 million," even though the precise figure differs meaningfully from the rounded one.
Rounding negative numbers, an edge case worth stating explicitly. The same digit-based rule applies regardless of sign — rounding −4.87 to one decimal place looks at the digit after the target place (7, ≥5) and rounds up in magnitude: −4.9, not −4.8. It's easy to instinctively round toward zero instead of following the digit rule strictly, which produces a wrong answer for negative values specifically.
For how rounding interacts with scientific notation and precision reporting specifically, see What Is Scientific Notation?