Definition
Accuracy is how close a measurement is to the true value. Precision is how close repeated measurements are to each other. The two qualities are independent: a measurement can be precise without being accurate, accurate without being precise, both, or neither. A bathroom scale that reliably reads three pounds too high is precise but inaccurate. A scale that bounces around the true weight by five pounds in either direction is accurate on average but not precise.
The dart-board analogy is the standard teaching device. Tight clustering anywhere on the board is precision. Hitting near the bullseye on average — even with a wide scatter — is accuracy. Tight clustering on the bullseye is both, and the goal of any well-designed measurement.
Why it matters
How it works
The conceptual split maps neatly onto two kinds of measurement error. Random error is the unpredictable scatter of repeated measurements around their average — it degrades precision but not accuracy. Averaging many independent measurements reduces random error in proportion to the square root of the sample size, because the noise tends to cancel. Systematic error is a consistent bias that pushes all measurements in the same direction — a miscalibrated instrument, a measurement procedure that always rounds in one direction, or a sampling frame that misses part of the population. Averaging more measurements does NOT help here; the bias persists no matter how many samples you take.
The practical workflow that delivers both qualities is calibration plus replication. Calibration anchors the instrument to a known reference — a certified weight, a primary time standard, a control sample of known concentration — eliminating or quantifying systematic error. Replication then beats down random error by taking many independent readings and averaging them. The risk in shortcutting either step is to mistake one quality for the other. Reporting many decimal places on a single uncalibrated reading is a classic confusion of precision for accuracy. The third significant figure of a wrongly-zeroed scale is meaningless. Conversely, a single calibrated reading without replication gives no information about the noise around the answer.