This paper deals with the development of a realistic power line channel simulator wherein power line communication devices can be tested before implementation to meet the massive need of data transfer. The statistics of the noise follow the experimentally observed in different sites, namely the time-varying non-white power spectral densities (PSDs) of the background noise and a target non Gaussian amplitude distribution. The procedure based on the inverse cumulative distribution function method of generation of random numbers and iteratively updating a target spectrum necessitates knowledge of a maximum of 17 parameters for successful implementation and has been validated for three sites in the low-frequency ([Formula: see text]500[Formula: see text]kHz) and high-frequency (1–30[Formula: see text]MHz) bands. The average percentage errors in prediction of the mean of the channel capacity (CC) are 12.68% and 10.66% in the two bands, respectively. The minimum correlations of the distribution of BER of OFDM in a channel corrupted by the simulated and observed noises are is 0.883 and 0.801 in the two bands which are high compared to 0.422 and 0.355, respectively, when the requirement of a target amplitude distribution is neglected. With low-frequency noise emulated by a data acquisition card, an average percentage error of 11.82% in the CC and a correlation of 0.867 (against 0.498) in BER are obtained. The noise thus generated can be used as a testbed for system testing, instead of the conventional static models (additive white Gaussian noise or with time-invariant colored PSD), leading to better optimization of the implemented devices.