A new kernel density estimation scheme which incorporates gray values, spatial relation and local standard deviations information effectively was proposed. With this scheme, an infrared target extraction method based on kernel density estimation was designed. Firstly, the entire image Was divided into blocks and the confined region that contained the entire target was obtained based on the statistical feature of each blocks. Secondly, the kernel density estimations of the confined region and the corresponding background sample region were estimated. Here, the background sample region is a surrounding image region of the target, but it doesnt contain the target. It is enlarged from the confined target region. Finally, the target was extracted with the contrast of two kernel density estimations under a given threshold. The experimental results show that the proposed method is simple and efficient.