In this paper, the impact of statistical noise modelling on the error performance achieved by orthogonal frequency-division multiplexing (OFDM) over indoor broadband power-line channels is investigated. Different classes of statistical models suitable to represent power-line noise are illustrated and their impact on the error performance of a specific OFDM system is assessed via computer simulations. Numerical results are compared with the error performance provided by the same system in the presence of measured power-line noise; this is evidence that a realistic indication of error performance can be achieved only if the power spectral density of the adopted noise model exhibits a good match with that of the measured noise. In practice, this result can be achieved by modeling the power-line noise as a moving average random process of proper order; however, a satisfying match can be achieved as well if other simple noise models available in the technical literature are adopted.