Abstract: Monte Carlo simulation (MCS) is widely utilized in the reliability evaluation of composite power systems while suffers from low computational efficiency. This paper proposes a data-driven reliability evaluation method based on the convolutional neural network (CNN). The CNN is adopted in the sequential MCS framework to improve its efficiency. In this paper, the features reflecting the operating state of the system are firstly constructed and the optimal load shedding model based on CNN is established. To solve the problems of unbalanced reliability evaluation samples and low training efficiency of regression model, the system state classifier is then established and the classification-regression model based on CNN is formulated. What's more, the correction mechanism of classification results is introduced to further promote the practicality of the classification-regression model. The validity and effectiveness of the proposed method are verified on the modified IEEE-RTS79 and IEEE-RTS96.