Neural networks with real valued inputs have been proposed in the literature for adaptive equalization and have been used to improve performance of communication channel equalizers. However, neural networks with complex valued inputs and fast convergence are lacking for adaptive equalization. Therefore, in this paper, complex extended Kalman filter (CEKF)-based neural network with complex valued inputs for adaptive equalization of a communication channel is suggested. Performance comparison of the CEKF and complex backpropagation (CBP) neural networks is made through simulation results.