A bootstrap procedure was used to determine the statistical precision of estimates of mean and peak spinal loads during lifting as function of the numbers of subjects and measurements per subject included in a biomechanical study. Data were derived from an experiment in which 10 subjects performed 360 lifting trials each. The maximum values per lift of the lumbar flexion angle, L5S1 sagittal plane moment, and L5S1 compression force were determined. From the data set thus compiled, 3000 samples were randomly drawn for each combination of number of subjects and number of measurements considered. The coefficients of variation of mean and peak (defined as mean plus 2 standard deviations) spinal loads across these samples were calculated. The coefficients of variation of the means of the three parameters of spinal load decreased as a linear function of the number of subjects to a power of about -0.48 and number of measurements to a power of about -0.06, while the corresponding powers for peak loads were about -0.44 and -0.11. © 2002 Elsevier Science Ltd. All rights reserved.