The performance of focusing evaluation function is an important factor that affects the on-board autofocusing of a scanning remote sensing satellite. Based on the imaging properties of the agile remote sensing satellite and computing characteristics of the on-board processing unit, an autofocusing strategy of the agile remote sensing satellite is designed, and an improved gray gradient function is proposed as the focusing evaluation function. Based on the traditional gray gradient function and the relationship between image edge information and defocus state, the proposed method takes gray standard deviation as the edge threshold to extract the effective edge information and calculate the image definition. The study results show that the proposed method has high sensitivity, strong unimodality and good timeliness, and is robust to the scene change and noise in the process of focusing imaging. Compared with other commonly used gray scale focusing evaluation function, the proposed algorithm shows obvious advantages.