Aiming at the need to establish a correct mapping relationship between input and output in solving inverse kinematics of robots by neural network, a method has thus been proposed of solving inverse kinematics of planar 2R manipulator based on BP neural network with unique feature. Based on a geometric analysis in the workspace of the manipulator, the law of multiple solutions of inverse kinematics can be determined. Meanwhile, azimuth angle is added to the input of BP neural network as the characteristic parameter. The simulation results show that the complete inverse kinematics solution of planar 2R manipulator can be obtained with only one BP neural network, with a prediction error between -0.002 rad and 0.002 rad. Keywords：inverse kinematics；multiple solution；planar 2R manipulator；BP neural network