The characteristics of macro and micro computer visions were analyzed,obtained results demonstrate that micro-vision blurred effects are caused by both geometrical and wave optics.A calibration experiment was carried out,which verifies the assumption of the linear relationship between spread parameter and step number.On the basis of above research,the existing autofocus technology based Depth from Focus(DFF) was introduced and a fast autofocus algorithm based Depth from Defocus(DFD) was presented in micro computer vision.The algorithm can calculate the focusing position of a target directly by giving two frames of blurred images.Experimental results indicate that the focusing speed obtained by proposed algorithm is 2-4 times faster than that gotten by traditional method(Sum Modified Laplacisn Method,SLM).It is shown that the proposed method has better focusing performance and robustness and can be used in the autofocus for micro computer vision systems.