Space optical system with short focal depth is affected by the launch and working environment, and it is prone to random defocus errors, resulting in a serious decline in image quality. To solve the problems that refer to slow speed and poor anti-noise performance in traditional auto-focus method, a auto-focusing method based on depth from defocus that can be used for remote sensing images is proposed. This method only obtains two defocused images from different positions, and through Spatial-Domain Convolution/Deconvolution transform, the defocus of the system can be estimated. In the case of no noise and 11.5dB noise within 17 times the focal depth, the average error of the estimated defocus amount is 7% and 10%, respectively. The imaging detector can be quickly positioned near in-focus position, and then accurately search. The result of experiment shows that, compared with the traditional depth-from-focus method, calculation of several frames of images compared with dozens of frames of images greatly improves the focusing speed. At the same time, this method has more accurate defocus estimation results and better anti-noise ability for remote sensing images.