The purpose of this study was to determine the reliability of average surface electromyography (EMG) within and between days. Nine subjects were tested on three different days. Five maximum and 10 submaximum (five at each of two different levels) isometric contractions were performed each day. The submaximum levels were determined by the 30% and 50% maximum voluntary contraction (MVC) on day one and kept constant over days two and three. The myoelectric signal from the triceps was full wave rectified and low-pass filtered at 1 Hz to yield a relatively noise-free average EMG that corresponded in time to the force signal. The scores were normalized to the force level. The reliability was estimated with intraclass correlations coefficients (R). The magnitude of score variation was also expressed as a ratio, coefficient of variation, CV = standard deviation/mean X 100%. R values were significantly greater for the submaximal levels than the 100% MVC; no significant difference existed between the two submaximal levels. The within-day CV values were similar for all three levels, ranging from 8% to 10%. The between-days variability was higher than that within days, ranging from 12% to 16%. The results suggest that submaximal isometric contractions are more reliable. Measurement error is substantial in this technique and should be included in statistical designs.