This paper addresses the issue of the effect of random error upon the estimates of above-or belowground net primary production (NPP). We show that ramdom errors in estimates of biomass do not compensate but always result in a positive bias in estimates of NPP. Second, we demonstrate that the larger the number of time intervals considered, the higher is the positive bias or overestimation error. An effect similar to an increase in the number of sampling periods is obtained by increasing the number of components utilized in estimating NPP. These are usually taxonomic or functional (depth, live, recent dead, etc.) components.We calculate the magnitude of the overestimation error as a function of the size of the difference in two sequential estimates of biomass and the variability associated with them. We propose a method that uses this error to correct the estimates of NPP for the positive bias resulting from random errors and to develop confidence intervals for the corrected NPP. We suggest that this method will remove the positive bias from estimates of NPP upon which nutrient budgets, energy flow and trophic webs rely. The concepts presented will help in the design of experiments that use production as a response variable.