The invention discloses a social network friend recommendation method based on community division. Link prediction is to predict the connection possibility between two points according to an existingnetwork topological structure, node property information and the like. Most existing node similarity algorithms only consider information of common neighbor nodes, that is, a topological structure with path length of 2, the important information that some nodes possibly belong to the same community is ignored, and obviously the nodes in the same community are more possible to have links. Accordingto the defects of traditional link prediction methods, an improved genetic algorithm is mainly used to perform community division on all the nodes first, then link prediction is performed according to the community division result, and therefore a social network friend recommendation algorithm based on community division is proposed. By doing contrast tests among five real networks, the accuracyof the algorithm compared with the traditional node similarity algorithms is analyzed through comparison, and the availability of the algorithm is proved.