Previous studies have demonstrated the acceleration signal presents a typical running signature, which allows for the extraction of reliable information. However, few studies have focused on the exhaustion-induced variability of the acceleration signature during running. The present study included 10 participants who ran at a constant speed on a treadmill until exhaustion. The participants were equipped with three accelerometers, located at the lumbar spine, tibia, and foot. The results showed that all the participants kept a constant pace throughout the test (coefficient of variation <5%). Similarities between acceleration signatures were observed using the coefficient of multiple correlation. For the longitudinal axis of the lumbar spine, the longitudinal axis of the tibia, and the anteroposterior axis of the tibia, running signatures were not affected by exhaustion (coefficient of multiple correlation >0.8). For all the other axes, the signature was impacted within and between the states of exhaustion. Signatures were particularly different for the foot sensors, which makes it difficult to use to extract reliable information. The results showed that the coefficient of multiple correlation allowed the quantification of the variability of the running signature, and that each axis and measuring point varied in how they were influenced by exhaustion.