Recording a single sweep visually evoked potential (VEP) results in a poor signal to noise ratio. This ratio can be increased by the use of a filter based on an autoregressive with exogenous input model. 12 diabetic and 14 normal subjects were examined using a pattern reversal stimulus running at 4 reversals per second, with a contrast of 75%. The screen subtended a visual angle of 16 degrees, the check size was 60 minutes and the mean luminance was 70 cd/m 2 . The model allowed for extrapolation of the P100 component from each single sweep of the VEP. The VEPs obtained by conventional averaging were not significantly different in the two groups studied but the single sweep analysis showed differences in the P100 amplitude and latency between the normal and the diabetic subjects. The new technique appeared to be able to extract information that was masked by the normal averaging approach.