We derive analytical expressions for the output spectral density and the noise power $P_{eta}$ in noise loading analysis using Volterra kernels to characterize fiber nonlinearities. The bandwidth of the input noise source has little effect on $P_{eta}$, but the power of the input noise source and the dispersion parameter value of the fiber have a significant effect on $P_{eta}$. The Volterra method predicts ${Delta}P_{eta}[dB]$ = 30 dB/decade, which agrees very accurately over a wide range of fiber parameters compared with the numerical results by the split-step Fourier method. Therefore the Volterra method could be useful to predict the performance of a dense WDM system when we plan to upgrade fiber or increase signal power.