We used a simulation approach to investigate the implication of two methods of forming composite samples to characterize segmented populations. We illustrate the case where the weight of individual segments varies randomly, a situation common with fish samples. Composite samples from segments such as whole fish or muscle tissue should be formed by homogenizing each segment separately and combining equal-sized portions randomly drawn from each homogenate. This approach permits unbiased estimation of the mean concentration per fish. Estimates of mean contaminant concentration varied little with variation in the number of composite samples analyzed or with composite size (number of segments in a composite sample). However, for a fixed number of composite samples, the precision of the variance estimate increases as composite size increased. In addition, for a fixed number of composites, the estimate of the variance stabilized as more segments were included in the composite samples. Estimates of the variance among fish or other population segments can be recovered using appropriate compositing procedures and specially-designed studies.