The fault diagnosis based on nonlinear spectral analysis is a new fault diagnosis theory. In allusion to the online estimation problem of generalized frequency response functions in this fault diagnosis method, a new method is studied for obtaining generalized frequency response functions of the detected system by using Volterra kernels in time domain. Based on the robust Volterra total least mean square adaptive identification algorithm, a new online estimation method for generalized frequency response functions is presented in this paper. The presented method can be used to efficiently reduce the online computation requirements of generalized frequency response functions, and is can be easily realized. The simulation results indicate that the presented method is efficient.