An improved "three steps" mountain-climb searching (MCS) algorithm is proposed which is applied to auto-focusing for microscopic imaging accurately and efficiently. By analyzing the performance of several evaluation functions, the variance function and the Brenner function are synthesized as a new evaluation function. In the first step, a self-adaptive step length which is much dependent on the reciprocal of the evaluation function value at the beginning position of climbing is used for approaching the halfway up the mountain roughly. Secondly, a fixed moderate step length is applied for approaching the mountaintop of the variance function as closer as possible. Finally, a fine step is employed for reaching the exact mountaintop of the Brenner function. The microscope auto-focusing experiments based on the proposed algorithm for blood smear detection have been carried out comprehensively. The results show that the improved algorithm can not only guarantee the precision to get clear focal images, but also improve the auto-focusing efficiency.