The fiber-reinforced polymers (FRP) bar is a promising solution to problems caused by steel rebar corrosion in concrete. To assess the service life of the FRP bar based on accelerated test results, it is crucial to have a reliable model. Here, a modified exponential (MEP) model is proposed based on the Avrami equation. The Avrami equation provides a theoretical foundation for the empirical exponential (EP) model and does not a priori fix the power of the exposure time to one. A database containing 903 data points from 74 groups of test specimens is assembled to compare the reliability of the MEP model vis-a-vis the EP, single logarithmic, double logarithmic, and power function models. The combination of Root Mean Square Error (RMSE), the Mean Absolute Error (MAE), and the coefficient of determination (R2) criteria is proposed for assessing model reliability. It is shown that in certain cases the combined criteria, versus R2 alone, significantly increase the number of test groups meeting the acceptable performance limit. Observed test data aberrations are found to have minor influence on the results of the EP model, but they significantly influence the results of the other four models. The EP model generally predicts the lowest activation energy and the smallest strength retention for similar groups of bars, while the predicted values of the other four models exhibit a relatively small difference. The difference between the predicted strength retention values of the EP and MEP models shows an increasing trend with the increase of the absolute value of (1 − n), where n is the power of the exposure time in the MEP model.