The invention discloses a channel correlation-based user group partition method under D2D. The method comprises the following steps: (1) enabling users to measure a pilot frequency transmitted by a base station, and feeding back channel correlation matrixes and receiving signal intensities by means of an uplink; (2) obtaining the estimated positions of all the users according to an MUSIC (multiple-signal-classification) method and an RSS (receive-signal-strength) method, and adopting a binary group to indicate user positions; (3) carrying out clustering processing on the user by using a K-means method. The user group partition method provided by the invention can complete the partition of a two-step pre-coded user group of a cellular network and the discovery of a D2D cluster at the same time by utilizing cellular channel correlation matrix information, thus avoiding a complicated operation of directly utilizing channel correlation information to process medium and high-dimension matrix subspace division and projection; the method has the characteristics of being simple and efficient.