Background: Neuroplasticity drives recovery of walking after a lesion of the descending tract. Intramuscular coherence analysis provides a way to quantify corticomotor drive during a functional task, like walking and changes in coherence serve as a marker for neuroplasticity. Although intramuscular coherence analysis is already applied and rapidly growing in interest, the reproducibility of variables derived from coherence is largely unknown. The purpose of this study was to determine the test-retest reliability and agreement of intramuscular coherence variables obtained during walking in healthy subjects. Methodology/Principal Findings: Ten healthy participants walked on a treadmill at a slow and normal speed in three sessions. Area of coherence and peak coherence were derived from the intramuscular coherence spectra calculated using rectified and non-rectified M. tibialis anterior Electromyography (EMG). Reliability, defined as the ability of a measurement to differentiate between subjects and established by the intra-class correlation coefficient, was on the limit of good for area of coherence and peak coherence when derived from rectified EMG during slow walking. Yet, the agreement, defined as the degree to which repeated measures are identical, was low as the measurement error was relatively large. The smallest change to exceed the measurement error between two repeated measures was 66% of the average value. For normal walking and/or other EMG-processing settings, not rectifying the EMG and/or high-pass filtering with a high cutoff frequency (100 Hz) the reliability was only moderate to poor and the agreement was considerably lower. Conclusions/significance: Only for specific conditions and EMG-processing settings, the derived coherence variables can be considered to be reliable measures. However, large changes (>66%) are needed to indicate a real difference. So, although intramuscular coherence is an easy to use and a sufficiently reliable tool to quantify intervention-induced neuroplasticity, the large effects needed to reveal a real change limit its practical use.