Small UAV detection has special significance in the field of anti-UAV and UAV control. Due to the change of object size, illumination condition, complex background and other factors, the detection results will be affected. A novel small UAV detection method is proposed based on visual saliency to overcome these problems. First, the histogram is used based contrast method to detect the visual saliency features. Second, the selective search method is used to acquire the propose region set. Then, the visual saliency features is used to optimize the propos region set. Final, the normalized gradient features is extracted for each propose region, and the SVM classifier to detection is used UAV objects. The simulation results show that the mean average precision of our method can reach 84. 74%, and it still maintains 80. 97% under the condition of 20 dB noise. It suggests that our detection model has good detection ability for small UAV, and has potential engineering application value.