Abstract: The detection of infrared small targets under the condition of complex background is one of the most important technologies and the most popular research topics in infrared auto target search and infrared guarding systems. In order to detect the small target efficiently, the local gray mean was proposed to determine the size and location of the infrared small target based on the analysis of the characteristics of the small target and its local background. Firstly, the necessary condition to judge whether a pixel was a small target was given. Secondly, based on the pixels which was the target, the possible sizes of the small target could be figured out. Then, the result was optimized and false alarm was eliminated as far as possible. Finally, according to the result obtained from the first three steps, the ultimated size and location of the infrared small target were given. Matlab simulation results show that, for the complex cloud backgrounds, the detection algorithm based on Top-Hat transform has good detection speed, but when the gray values of false alarm and target are equal, the algorithm could not detect the target efficiently. When choosing suitable parameters, the new algorithm not only can accurately give the location information of the small target, but also has good ability to estimate the small target size. However, the detection speed of the new algorithm remains to be further improved.