The correct budget allocation for road maintenance, which represents a significant infrastructure investment in urban roads, requires the accurate prediction of the deterioration of bituminous hot mix asphalt (HMA). In this study, three different deterioration models have been developed that can predict the future performance of pavements in urban HMA paved roads. First, the current condition of the pavements was measured by using the pavement condition index (PCI), which is approved by the PAVER system. Then, three different models were developed to predict deterioration in the PCI as a function of pavement age, and applied to urban road networks in Samsun (Turkey). The models used were deterministic regression analysis, multivariate adaptive regression splines (MARS) and artificial neural networks (ANN). Variations of each model were explored and the one with the highest computational efficiency was employed for ranking pavement sections with respect to rehabilitation needs. Results indicated that the three approaches had comparable prediction accuracies and R-squared values, although predictions provided by the ANN model were more accurate compared with the other models. The article provides a detailed comparison of the performance of the three models. (c) 2016 Elsevier Ltd. All rights reserved.