Abstract: The research of ranging based on received signal strength indication (RSSI) has been a hot spot in the Internet of Things due to its wide application. In order to reduce the traditional RSSI ranging error based on back propagation neural network, a back propagation neural network RSSI ranging method based on K-means clustering algorithm preprocessing sample data is proposed. The algorithm solves the problem of non-uniform mapping between RSSI value and real distance in different distance intervals due to different attenuation degree of RSSI value. The K-means clustering algorithm is applied to BP neural network model to divide the distance interval of the sample data, and then the classified data are input into BP neural network to build the network model and simulate the experiment. The results show that the root mean square error of the traditional RSSI ranging method based on BP neural network is 1.4257 m, while the root mean square (RMS) error of BP ranging method neural network improved by K-means algorithm is 1.2887 m. Compared with the existing ranging methods based on BP neural network, it reduces the ranging error and optimizes the mapping relationship between the target RSSI and the real distance.