In the fault diagnosis domain for aero-engine, the data collected has the properties of high-dimension and nonlinear. Traditional pattern recognition method is difficult to learn the essential information of such data, which leads to the low fault diagnosis rate. Therefore, mining the feature reflecting aero-engine's condition from high-dimensional data is of great practical significance. To this end, a fault diagnosis algorithm using Stacked Denoising Auto-encoders(SDAE) for aero-engine is proposed in this paper. This method firstly map the high-dimensional data into low-dimensional features to extract feature by constructing SDAE from raw vibration signals of different conditions; Then a softmax regression model is used to verify the discriminability of the features. Finally, the validity of the method is verified by the experiments which show that the approach proposed is effective for aero-engine fault diagnosis.