Construction, maintenance and rehabilitation of highway infrastructures are the most essential issues for developing and improving a sustainable transportation industry. Besides high costs of these issues, improper performance of some aspects such as safety and security creates some problems which will threaten the reliability of a transportation system. Crash prediction models (CPMs), as safety analysis tools, have a specific role in analyzing the contributing factors in occurrence of traffic accidents. This study aimed to evaluate the effects of pavement distresses on the frequency of run-off-road accidents which are the most frequent and severe accidents in Iran and around the world. In doing so, about 150 km of Semnan-Tehran divided rural highway was examined and its pavement condition indice (PCI) was calculated. Then, using GENMOD and NLMIXED procedures in SAS software, two generalized linear models and 39 nonlinear negative binomial regression models were developed. After evaluating the goodness-of-fit of the models and estimating their error, comparison of these models leads to the most proper model. To estimate the goodness-of-fit of the models and select the best-fitted model, several criteria were implemented including standard deviation/scaled deviation, Pearson χ2, -2LL, AIC, AICc and BIC. Moreover, two criteria of mean absolute error and root mean squared error were used to estimate the validity of model results. In analyzing of residuals by standardized residuals method, the mean error was about -0.1048 and the variance was 0.9259, which are close to the assumptions of zero mean and variance equal to 1. CPMs showed that nonlinear negative binomial regression model fits significantly better with the available data, which demonstrates the association between PCI and frequency of run-off-road accidents. Sensitivity analysis of the model indicates that for nonlinear CPM, a unit increase in PCI results in around 1.93% reduction in frequency of run-off-road accidents.