Through the generalized additive models used in this study, several key points emerged. First, development stage was the single most predictive factor in the models assessed, explaining 89.5 percent of the deviance in the data. In contrast, the nonlinear measurements, weight and length, proved far less effective in predicting development; strain and temperature (genetic and environmental factors) were not significant predictors by themselves. Second, error increased as development progressed for all models, indicated by the gaps in predictive ability and the widening confidence intervals for successive developmental stages. Third, the limited influence of fly strain and rearing temperature on development was found to be an important consideration, since it indicates the models have value regardless of where flies are collected or at what temperature they develop. Fourth, any given forensic case may present the entomologists with different data from which to estimate fly age. Although developmental stage was the most useful datum for the development estimates in this study, other data, such as weight and length, increased their accuracy. Finally, and most important, a comparison of modeled development predictions to the independently derived rat data made it possible to assess error rates and produce confidence intervals in these estimates. The utility of the methodology presented is that it establishes a defined way of producing confidence intervals associated with entomologically based PMI predictions, regardless of fly age. 1 table, 4 figures, and 17 references