Abstract:The methods of multi-step grey-correlation evaluation and fuzzy comprehensive evaluation are usually used in dam safety evaluation. Both of them need to design weights of each index and membership of evaluation indexes,and then the dam safety is determined synthetically. Due to the complexity and diversity of each problem,the subjective factors have great influences on the final evaluating conclusion. The artificial neural network could adjust the weights of each influencing factor automatically. It could not only absorb the experts¢ thought and experiences embodied in the study samples,but also possess high anti-jamming ability and better error permissibility. As a result,an improved BP neural network has been applied to comprehensive evaluation of dam safety. However,the shortcomings of slow convergent rate,poor stability and local minimum of BP neural network have extremely restricted its application. Therefore,the radial basis function neural network is proposed to apply to comprehensive evaluation of dam safety. By study of the given samples,the experts¢ knowledge,experiences,subjective judgment and tendency towards the importance of objectives embodied in the samples are obtained. When applying the well trained network to map a new input of given samples,the output results can reproduce the experts¢ instinctive thought and experiences and make a reasonable evaluation conclusion. Evaluation examples of ten typical dam sections of Fengman dam testifies the validity of the new method.