The invention discloses an advertisement recommendation method based on a feature cross joint deep neural network. The method comprises the steps that 1) a server collects advertisement logs of an advertisement platform for data cleaning; data samples are added into a sample stream; data are stored in a storage module of a distributed file system; 2) the server uses a recall layer to filter the data of the sample stream to acquire preliminary candidate recommendation advertisement ID subsets for a user; and 3) the server sorts and predicts the candidate recommendation advertisement ID subsetsto acquire a corresponding user advertisement push subset. According to the invention, the effectiveness of advertisement recommendation is improved, and an advertisement CTR index is improved.