A real-time, accuracy and robustness multi-resolution mountain climb servo (MCS) auto focus method was proposed in order to solve real-time auto focus problem, and was obtained high clear high quality image because low clear low light images was difficult to realize auto focus in range-gated imaging system for underwater. The merits of this method were outstanding as follows. It reduced the amount of processing data and time by taking advantage of the image multi-resolution theory by spatial sampling of the current image and then formed different resolution sampling layers, which made the system possible to deal with auto focus more quickly as 1/2 time as traditional method. Thereby, the method calculated current frame weighted average edge detection image definition, which was determined by the correlation between current frame and two adjacent frames with 0.5, 0.3 and 0.2 coefficients before solving the error judge of interference maximum local peak. At last, the method realized the last operation auto focus by MCS. The repeated experimental verification of range-gated imaging system for underwater shows that the algorithms is capable of guaranteeing the auto focus speed and accuracy and improving the real-time, accuracy and robustness in focus determination.