Spectral EMG parameters are being used as an objective evaluation of low back rehabilitation programs. The reliability of these spectral parameters is important in determining the validity of this evaluation tool. Two groups of eight subjects, with no history of back pain, were measured: the first group every day for one week; the second group one day per week for four weeks. During each session, subjects performed a 30-second isometric fatiguing contraction of the back extensors at 60% MVC followed by a 60-second rest and a 10-second repeat contraction. The intraclass correlation coefficient (ICC) did not demonstrate high reliability (ICC < 0.6) for the rate of median power frequency (MedPF) change (Hz/s), the magnitude of the recovery (Hz) and differences between the dominant and the non-dominant sides within a subject. However, the initial MedPF (Hz) of the fatigue and repeat contractions demonstrated excellent reliability (ICC > 0.8) with five or more repeated measures. The practical implication of this work is that the experimental condition must have sufficiently large changes in MedPF (signal)--at least larger than the variability inherent in the MedPF (noise)--to constitute a valid measure.