In this paper, a new method for radio wave propagation prediction based on BP neural network and simplified path loss model is proposed. The method is based on a large amount of the Minimization of Drive Test (MDT) data generated before the antenna adjustment, which include reference signal receiving power (RSRP). And by fitting radiation pattern, the simplified path loss model is used to get the RSRP after the antenna adjustment. In the proposed algorithm, first, the cell is divided into multiple grids. Then the BP neural network is trained by using the RSRP means of grids with sufficient samples before the antenna is adjusted. Next, the trained network is used to predict the RSRP means of grids with insufficient samples. After the antenna adjustment, the predicted RSRP mean of each grid is obtained based on the simplified path loss model. In order to verify prediction accuracy, we recalculate the statistical mean and standard deviation of the RSRP samples in each grid based on the MDT data reported after the antenna adjustment. Two LTE base stations were tested. The results show that the method makes an accurate prediction for RSRP mean of the grid with less cost.