Time series models developed by regression to forecast highway crash rates in association with some selected special population subsets are described here. Such models are useful for assessing the safety performance of the subsets because the problems inherent in subsets are different from those of the average highway population. They also enable high-risk groups to be identified, not only under present conditions but also in the future. Two sets of models were developed: one for the state of Florida and the other for United States. The population subsets considered included older drivers, young drivers, international tourists, and school-age children as nonmotorists. Among the different model formats tried, the negative exponential was the best fitting and gave reasonably good fitness, as indicated by considerably high R² values varying from 0.784 to 0.974. The main purpose of the model building was to forecast the crash rates by each population subset at various time points within the study horizon. Previous studies by other researchers where forecast values were compared with actual values have indicated that most of the complex models were incapable of explaining the future situation. As such, the opinion of the authors is that simple models with year as the independent variable could be advantageous for forecasting the safety performance of the highway system for special population groups.