To compare with radio communication, free space laser communication is owning to higher data rate, higher security and less power dissipation, has become one of the research hotspot. However, the performance of free space laser communication is limited to some factors, such as signal fluctuation, the difficulties of tracking and pointing mechanism, and the wavefront aberration caused by atmosphere turbulence. In order to improve the performance in complex environment, the deep learning method is introduced to free space laser communication. Various studies have shown that the deep learning method can process signal with obvious advantages in many aspects of free space laser communication. This paper makes a review of the deep learning method used for free space laser communication signal processing and detection, the tracking and pointing mechanism, and wavefront aberration detection and correction. Finally, the prospect of deep learning technology for FSOC is prospected.