A new channel estimation method of neural network based on complex radial basis function (CRBF) is proposed to enhance the anti-interference ability of traditional pilot frequency estimation algorithm in power line communication (PLC). This method builds up a new channel model of complex field signals in PLC. The complete response model was established by using transmitting terminal's pilot signal as input sample data, pilot signal's frequency response as output sample data, and pre-setting mean square error (MSE) and diffusion constant. Computer simulations show that compared with the traditional algorithm the channel estimation was more accurate and had lower MSE and bit error rate (BER).