Aimed at the problems that the traditional fault detection and diagnosis methods based on fixedthreshold have a high false alarm rate and fail to realize effective fault detection and isolation, an adaptive fault detection and isolation method for hydraulic servo system based on multistage observer is presented. The firststage RBF neural network is adopted as a fault observer of the hydraulic servo system, and the residual error signal is generated by comparing the estimated observer output with the actual measurements. The secondstage RBF neural network is employed as an adaptive threshold producer, realizing the adaptive fault detection. Features of the residual error signal are extracted by using wavelet packet analysis, and the system fault isolation is made by using the thirdstage RBF neural network. The experimental results show that the multistage observer is effective in detecting and isolating the failure in the hydraulic servo system.