AbstractSafety performance functions (SPFs) are used in roadway infrastructure safety management for purposes such as crash contributing factor analysis and hazardous roadway section identification. However, due to the lack of traffic data to establish SPFs particular to their locality, some regions and cities need to transfer and calibrate SPFs from other jurisdictions. In China, Shanghai City has been the only city that has conducted major safety analysis and developed facility-specific SPFs, while other cities have not built their local SPFs for lack of sufficient reliable data. To assist most cities in developing safety analysis models quickly enough to keep pace with the increase in crashes, the transferability of SPFs from Shanghai to another Chinese city of Guangzhou was explored. Considering the potential correlations between crash frequencies and explanatory variables during different times of day, negative binomial models were developed separately during morning peak-hours and off-peak-hours. When the models were evaluated by cumulative residual plot, results suggested that the local models could not be transferred directly without proper calibration. To provide insights into the selection of calibration methods considering the goodness-of-fit and practicality, this study compared various calibration methods that have been used in the literature and recommended a full Bayes method with normal distribution as informative priors to determine their practicality. The performances of calibration factor and calibration function, pooled data, the empirical Bayes method, and the full Bayes method were evaluated by mean absolute error, mean squared error, transfer index, and cumulative residuals plot. Among the tested methods, the calibration function method and the full Bayes method are recommended since they perform well in prediction accuracy and require an adaptable sample size.