Traditional gradient based image sharpness evaluation functions are sensitive to noise, which is easy to introduce more non-edge information and shows poor stability in practical applications. This paper analyzes the effect of the defocus of an optical imaging system on image sharpness based on the defocused model and proposes a improved sharpness evaluation method. The method uses the maximum gradient as the criterion of image sharpness.Then,it introduces the optimal threshold to distinguish the edge points and non-edge points to reduce the impact of the gradient of flat area and the noise in the image on evaluation results and to suppress the noise interference. The simulation experiments and actual test are performed and obtained results are compared with that of different traditional methods. It is verified that the improved method has better sensitivity and noise immunity, which is able to evaluate the sharpness of defocused image accurately and is suitable for auto-focusing in actual optical imaging systems.