A major critical challenge for hill climbing algorithm is the problem of poor accuracy and slow speed in the automatic focusing. Here, a partition search algorithm combining self-adaptive step and two-step search are therefore proposed to solve this challenge. This method divides the focus evaluation curve into gentle and steep region according to the value of focus evaluation function and curve shape. In gentle region, self-adaptive step is utilized to increase the auto-focus search speed. Then two-step search method is explored in steep areas to avoid focusing errors caused by false peak interference and improve auto-focus accuracy. The experimental results show that the focus speed of the partition search algorithm is increased by 11.25% compared with the hill climbing algorithm, and it has better accuracy and anti-interference ability.