Innovative applications of non-linear time series analysis have recently been used to investigate physiological phenomena. In this study, we investigated the feasibility of using the correlation integral to monitor the localized muscle fatigue process in the biceps brachii during sustained maximal efforts. The subjects performed isometric maximum contractions until failure in elbow flexion (90 degrees from neutral). The median and the 70th percentile frequency of the Surface electromyography (SEMG) power spectrum, the integrated SEMG, and the Correlation Integral (CI) were evaluated during the trials. The linear correlation between these variables and the elbow torque production was used to quantify the ability of a parameter to follow the fatiguing process. The CI had the highest linear correlation with torque (0.77 (0.12SD)), while the spectral indices correlations with torque were much lower. The decreasing trend of the torque production was followed by the spectral indices for only the beginning part of the contraction, while the CI increased sharply after the torque production fell to about 0.60 of the MVC. This suggests that the CI is sensitive to different changes of the SEMG signal during fatigue than the spectral variables.