The state of Michigan has 191 432 km (118,950 mi) of roadway (paved and unpaved), including highways, roads, and streets. Local government agencies, which are responsible for 176 270 km (109,529 mi) of these roads and streets, commonly use a pavement management system (PMS) called RoadSoft to assist in managing their pavement network. A key element of any PMS is its ability to predict future pavement performance. A study is described in which various deterministic and probabilistic models were evaluated using data from two Michigan counties. It was found that the logistic growth model and the Markov model provided the best combination of predictive ability and potential for applicability in Michigan counties. A comparison between these models found that their predictive ability for four pavement segments with different deterioration rates was good, with the Markov model offering the added advantage of representing future performance as a probability distribution, not as a single condition state. Current plans are to implement the logistic growth model in RoadSoft by the end of 1999 and to add the Markov model as local organizations gather sufficient pavement condition data over the next 5 years. It is hoped that these two pavement deterioration models can be implemented in the RoadSoft PMS to improve pavement performance prediction.