Aimed at the problems that aero-engine is complex in structure, severe nonlinearity of various degenerate state is variable, and traditional physical failure model-based method is difficult to predict the remaining useful life of the engine (Remaining Useful Life, RUL) accurately, the problems abovementioned can be done by adopting an improved convolution neural networks (CNN). A linear degradation model is employed to label each sample. The convolution is set to several different onedimensional convolutions to extract data features and the correlation between the RUL better. In order to validate the effectiveness of the method, a test is made on the commercial modular aeropropulsion system simulation (C-MAPSS) aircraft engine datasets provided by NASA. The results show that the convolutional neural network has higher precision compared with the common neural network.