In the selection of an economical treatment for rehabilitation of a deteriorated pavement section, decision makers usually encounter various situations. Factors affecting selection of a flexible pavement maintenance strategy may include distress conditions, traffic volume, and road class, among others. Traditionally engineers make their selection on the basis of their experience and judgment and past maintenance data. Experts' judgments are usually compromised in a group discussion to construct decision trees or decision matrices or even to develop knowledge-based expert systems. An artificial neural network is known to be an efficient technique for selection of an appropriate maintenance strategy. A genetic adaptive neural network training algorithm with a single hidden layer and sigmoid squashing function constitutes the network. The input vector represents the factors affecting maintenance strategy selection, and the output vector represents the appropriate maintenance strategy. A set of examples is derived from experts' judgments with a total of 144 cases randomly divided into "in-sample" and "out-of-sample" data for training and testing purposes, respectively. The trained network successfully predicted 83% of the test cases. The remaining 17% of cases were one or two levels away from the expert judgments used in network testing. Neural networks provide an efficient and optimum solution for such complex problems with the added advantage of faster implementation and easier updating than with other traditional techniques.