The nature of the residual background noise in ABR averages was empirically examined in normal hearing objects. The residual noise in the average was estimated with use of the technique described by Elberling and Don [Scand. Audiol. 13, 187-197 (1984)]. Low-level click stimuli were presented in 2-dB steps spanning the range from 30 to 48 dB p-p.e. SPL. For each stimulus level, 10,000 sweeps were acquired and stored for analysis. Shortcomings of the use of artifact rejection and standard averaging are demonstrated. It is further demonstrated how application of the Bayesian estimation technique of Elberling and Wahlgreen [Scand. Audiol. 14, 89-96 (1985)] to form weighted averages can help minimize these shortcomings. Finally, the effects of smaller sweep block sizes on the Bayesian technique's ability to control the destructive effects of nonstationary noise are analyzed. Minimizing the destructive effects increases the value of statistical techniques used to detect objectively or to control the quality of ABR recordings. In all, these techniques in combination improve not only the accuracy of test interpretation but also the efficiency of clinical test time, which is becoming important for the control of medical costs.