The bigger the undulation of gray-level in a neighborhood region, the bigger the sum square of the ratio of each gray-level to the total gray-level in the neighborhood is. Thus a method of detecting smallmoving targets based on the distribution of gray-level in the neighborhood is described. Considering the complex background, especially which involves a lot of edges and high-frequency motionless points, we brought forward improved methods for detecting possible targets.Finally, the neighborhood judging method is presented to pick out the moving targets. Simulation result indicates that this approach can reduce the number of possible targets and detect small moving IR targets in complex natural background with high efficiency and veracity. Furthermore, it can be used in real-time and multi-targets detection.