A smoke detection algorithm combined of multiple motion features was presented. The main purpose of this paper was to separate the smoke-resembling natural phenomena such as clouds, cloud shadows and dust from real smoke. The optical flow property, correlation property based on discrete wavelet transform and diffusion property were matched together to make discrimination. At last, a data weighted fusion algorithm based on statistics theory was applied into smoke detection. Experimental results showed that, compared with the algorithm of detecting smoke using single property, the proposed method can improve the accuracy of smoke detection.