Video key frame extraction has been an important issue for video index and retrieval. In this paper, we propose a key frame extraction method based on the temporally maximum occurrence frame (TMOF). We took the hybrid projection value of image as the feature vector of the frames, and used the distance between the vectors to fragment video shot into sub-shots. Then, we constructed TMOF and selected the frame which was with the smallest distance from TMOF in the sub-shot as the key frame. The experiment results showed that the algorithm extracts the key frames which meet the shot semantic well and with high accuracy and low algorithm complexity.