Abstract We describe a method for determining optimal sampling designs for experiments involving subsampling in stages. An optimal sampling design either maximizes the precision of the results for a given amount of experimental effort (time, money) or minimizes the effort required to achieve a specified level of precision. The optimization method utilizes information on (i) the cost of obtaining a sample at each stage and (ii) the variability of the data attributable to each sampling stage. To illustrate the method, we use effluent suspended cell data from a packed-bed bioreactor. The sample design involved three stages. First, three primary samples were collected from the reactor. Second, each reactor sample was subsampled in triplicate and a specified volume of each subsample was passed through a filter. Third, the bacteria of the filter were counted in each of ten microscopic fields, each field representing the same proportion of the filter area. The optimization analysis, applied to the mean of the log-transformed bacterial counts, showed that the precision of the sample mean could be substantially improved by expending more sampling effort on the first stage and less on the second and thir stages. We also show how the optimization analysis can be applied to other response measurements (e.g., total organic carbon, soluble organic carbon and glucose concentration) and sampling designs involving other numbers of subsampling stages.