This study aimed at finding the effects of road geometry and cross-section variables on numbers of accidents. In addition, a methodology to combine variables by using decision trees was developed. Combination variables for road geometry and cross section were created by using the chi-square automatic interaction detection algorithm. Two negative binomial models were developed: one with homogeneous road segments and the other with 1-km road segments. Homogeneous road segments were divided on the basis of the horizontal alignment of the road. They were either curved or straight. The accuracy of the negative binomial model with homogeneous road segments was compared with that of the negative binomial model with 1-km road segments. The negative binomial model using homogeneous road segments was found to be the more accurate of the two models. The model with homogeneous road segments was used to draw conclusions with regard to the effect of variables on the number of accidents. Combination variables showed a significant effect on the number of accidents. The road geometry and cross-section variables were found to affect the number of accidents differently under various combinations of other variables.