Studies on the motor system commonly use averages of rectified electromyogram (EMG) to measure muscular response. The assumption is usually made that this is a linear measure of response magnitude. It is shown here using a theoretical and experimental model that averages of rectified EMG can lead to non-linearities, which are of particular importance when considering the summation of two independent responses. When responses had a highly stereotyped waveform from trial to trial, so that averages of unrectified EMG revealed a deflection from zero, in measurements from averages of rectified EMG the size of a response to two stimuli delivered together was greater than the linear sum of the responses to each stimulus delivered alone. This deviation from linearity was greatest when two responses which were small relative to the ongoing background EMG activity were combined. The total response was then as much as twice as large as the linear sum of the individual component responses. Conversely, under conditions where responses were highly variable from trial to trial, resulting in no consistent deflection being seen in an average of unrectified EMG, the summed responses in rectified and averaged EMG were smaller than expected if summation were linear. This effect was most pronounced when both component responses were relatively large; the combined response tended asymptotically to approximately 70% of the linear sum of the component responses. A practical method is presented which allows prediction of the size of a response to two stimuli given together, when measured from averages of rectified EMG, on the assumption that they act independently.