Abstract: For nonlinearity, disturbance, complexity and parameter uncertainty in the electro hydraulic position servo system, a fault detection method for electro-hydraulic position servo system based on kernel regression interval model was proposed. The method made full use of the measurable input-output data to construct interval model, where those data was capable of reflecting the uncertainties of electro-hydraulic position servo system in the absence of failure, the accuracy and structure complexity of the constructed interval model could be trade-off. The interval model was used to judge whether the actual output of the electro hydraulic position servo system exceeded the interval output to realize fault detection. Because this method considered the dynamic threshold based on interval for fault detection, it could eliminate the false alarm of fixed threshold and was capable of improving the adaptability in detecting fault. The experimental examples were investigated to demonstrate the effectiveness of the proposed fault detection method, which could be applied for the real-time fault detection of the electro-hydraulic position servo system.