The aim of this paper is to propose a method of data processing for the analysis of evoked potentials, in particular for auditory brainstem responses. The present method has been developed to simplify and speed up the interpretation of the recordings by means of an enhancement of the response peaks. Even for experienced observers, identification of the response waves and subsequent latency measurements may sometimes constitute a difficult task, due to the presence of residual noise or to interference between the temporal waveforms of adjacent peaks and troughs. The method is implemented with a digital non-causal (zero-phase shift) filter, based on the convolution with a finite impulse response, to make the computation time compatible with the use of low-cost microcomputers. The performance is shown to be very good in several examples.