A two-dimensional direction-of-arrival (2D-DoA) estimation approach based on radial basis function neural networks (RBFNN) is proposed in this paper. With the spatial cone angle, two RBFNN estimation models of two line-arrays of L-shape array are built respectively, which can estimate the spatial cone angle, namely dimension-degraded model. The intersecting line of two half-conical surfaces,which is corresponding to each spatial cone angle, is the arrival path of the unknown signal. Simulation results show that the proposed method can effectively reduce the training set, and the complexity of model building, it also has very high resolution, and effectiveness for future application.