In this paper, normalized difference cloud moving index (NDCMI) is put forward by analyzing the remote sensing data every 15 minutes. By applying NDCMI, the detection of low cloud, thin cloud and the edge of moving cloud can be enhanced. We proposed a dynamic threshold method with NDCMI for preliminary cloud detection, and then through the preliminary results, we proposed a novel cloud detection method based on back propagation neural network (BP-NN). The experimental results show that the proposed method can eliminate the subjectivity in threshold selection, and can achieve better cloud detection results for remote sensing image in complex situations.