In order to reduce property loss and casualties from level crossing accidents, it is crucial to develop effective accident prediction models that are capable of providing effective information of accident frequency and severity given a vector of covariates. In the present research, a set of statistical count and categorical data models are developed; they are not only able to evaluate accident frequency and severity but also capable of exploring the potential risk factors that are responsible for traffic accidents. Using the data set collected by the Ministry of Transportation and Communication (MOTC) in 1998, which consist of both historical accident data and railway level crossing related data, the empirical study identifies a vector of factors that are significantly associated with accident frequency and/or severity. Finally, the developed accident frequency and severity models are also employed to provide the evaluation of black spots and countermeasure effects.