The auto-focusing precision of microscopy has great influence on the performance of machine-vision-based precise measurement. A comprehensively quantitative evaluation method on image auto-focusing technique in a microscopic vision environment was researched. Several kinds of evaluation indexes were proposed, and the unbiasedness, unimodality, spatial resolution etc. of 13 groups sharpness functions were comprehensively evaluated in a microscopic vision condition. Then variance function and Brenner function were chosen to calculate the sharpness functions in coarse and fine focusing processes respectively. A modified Mountain Climbing Searching (MCS) algorithm was proposed to implement the micro-automatic focusing. As comparing to common MCS method, the modified method significantly improves the time consuming and increases the repeatability by about 24%. Finally, the developed auto-focusing algorithm was integrated into the system and was applied to the measurement of armature gap in a servo solenoid valve. The results show that the standard deviation of measurement is 1.9 μm, the precision is similar to that of the universal tool microscope, and the efficiency is significantly improved. Moreover, the system was also utilized for dynamic characteristic detection of gaps in the solenoid valve under the condition of power up, the relation between driving current and armature gap is obtained, which provides a reliable evidence for in-situ micro-assembly.