The nature of experimental error in the determination of initial velocities of enzyme-catalysed reactions was investigated. No support was found for the generalization that such velocities should be homogeneous in variance. Instead the variance increased with the velocity in all of the cases studied. The weighting of sets of replicate observations according to their sample variances was tested in simulated experiments and found to give unacceptable results. It was practicable to decide on the best weighting scheme by examining the variability of the deviations from fitted curves, provided that the number of observations was large enough.