This case study aimed to examine the spatial and temporal transferability of safety performance functions (SPFs), which were developed for the Iowa interstate system in the form of negative binomial regression models. Four years of crash data were integrated with geometric and traffic information over a four-year period for the entire interstate system. The segments were randomly split into two groups and these groups were also split into a pair of two-year time periods. This allowed for an assessment of model transferability across space, time and both dimensions. Separate models were estimated for each of the four datasets and these models were used in cross-validation exercises. The predicted values were directly compared to actual observed values to assess predictive accuracy. In this setting, less spatial variability was shown as compared to temporal variability, which was largely reflective of significant reductions in crashes that occurred over the course of the study period. In all settings, temporal transferability was relatively poor. The results improved when calibration exercises were conducted. Ultimately, the results support prior research, which suggests state-specific SPFs are recommended in order to obtain better predictive capabilities. Full models, which considered numerous predictor variables, performed better than simple models considering only annual average daily traffic. Additional research is suggested in this area in order to better understand how spatial and temporal transferability compares across different empirical settings.