Many authors produced carrion insect development data for predicting the age of an insect from a corpse. Under some circumstances, this age value is a minimum postmortem interval. There are no standard protocols for such experiments, and the literature includes a variety of sampling methods. To our knowledge, there has been no investigation of how the choice of sampling method can be expected to influence the performance of the resulting predictive model. We calculated 95 % inverse prediction confidence limits for growth curves of the forensically important carrion flies Chrysomya megacephala and Sarconesia chlorogaster (Calliphoridae) at a constant temperature. Confidence limits constructed on data for entire age cohorts were considered to be the most realistic and were used to judge the effect of various subsampling schemes from the literature. Random subsamples yielded predictive models very similar to those of the complete data. Because taking genuinely random subsamples would require a great deal of effort, we imagine that it would be worthwhile only if the larval measurement technique were especially slow and/or expensive. However, although some authors claimed to use random samples, their published methods suggest otherwise. Subsampling the largest larvae produced a predictive model that performed poorly, with confidence intervals about an estimate of age being unjustifiably narrow and unlikely to contain the true age. We believe these results indicate that most forensic insect development studies should involve the measurement of entire age cohorts rather than subsamples of one or more cohorts.