Because of the combined practical and theoretical interest in pavement performance modeling, much research has concentrated on this area during the past few years. The objectives of this study are, first, to cluster pavement distress data into homogeneous groups depending on the extent of distress and, second, to identify and assess the influence of a variety of exogenous variables on the extent of pavement distress. To achieve these objectives, a two-step clustering and classification evolutionary modular neural network is developed with a large (more than 1,000 observations) and recent (1998) data set collected from in-service pavements in 15 European countries. The results indicate that the extent of the cracking distress (magnitude of cracking) can be separated into five homogeneous clusters (with Cluster 1 suggesting very low cracking and Cluster 5 very extensive cracking), with different explanatory variables affecting the extent of cracking in each of these clusters.