Mean values of stream compositions in mineral processing plants are frequently estimated by the analysis of a composite sample. This sample is the result of an accumulation of material increments extracted at systematic time intervals. The measurement error which is evaluated in this study is the difference between the sample composition and the true composition of the total material which flows during the time period of increment accumulation. The error variance is calculated assuming that the species flowrate random variations are described by their variances and autocorrelation functions. The error is decomposed into two contributions: the discretization error and the sampling error. The results are compared to those of Gy＇s theory of particulate material sampling. This estimation of the measurement error variance is essential for the application of data reconciliation or filtering techniques, for gross error detection, and for the estimation of process performance indicator reliability.