A technique for cancelling background noise in sensory evoked potential recordings is described. Since sensory evoked potentials are generally much smaller than the background noise, a large number of individual responses must be averaged. The proposed technique uses a second recording channel to estimate the background noise; a cancellation filter function is derived for each individual response in the average. The filter can vary from response to response; however, it does not attempt to adapt to long-term changes in the noise. The technique is intended for situations where the noise is intermittent or rapidly varying, as is often found in the operating room and intensive care unit. The technique was evaluated by a simulation using a known brainstem auditory evoked potential signal plus EEG samples obtained from a patient in the intensive care unit during an evoked potential test. The results showed that the technique could provide improvements in signal-to-noise ratio of 3-6 when the dominant noise component was external electrical interference. The effect was less when the background noise was muscle activity or spontaneous EEG, but there was still significant improvement. It was also found that the improvement obtained was approximately equal to the improvement that could be predicted from the formulation of the filter function.