A decision method for pavement maintenance based on neural network is proposed. This method is used to compute different evaluating indicators as Pavement Condition Index (PCI)，Structure Condition Index (SI)， Riding Quality Index (RQI) and Annually Average Daily Traffic (AADT) in terms of their membership function combining with various classical strategies which aims at improving the feasibility of these indicators' performance. After that，a neural-network decision model of pavement maintenance is established according to these evaluating indicators in order to obtain an optimal decision of pavement maintenance.