Near-infrared spectroscopy (NIRS) is a common tool used to study oxygen availability and utilization during repeated-sprint exercise. However, there are inconsistent methods of smoothing and determining peaks and nadirs from the NIRS signal, which make interpretation and comparisons between studies difficult. To examine the effects of averaging method on deoxyhaemoglobin concentration ([HHb]) trends, nine males performed ten 10-s sprints, with 30 seconds of recovery, and six analysis methods were used for determining peaks and nadirs in the [HHb] signal. First, means were calculated over predetermined windows in the last 5 and 2 seconds of each sprint and recovery period. Second, moving 5-seconds and 2-seconds averages were also applied, and peaks/nadirs were determined for each 40-seconds sprint/recovery cycle. Third, a Butterworth filter was used to smooth the signal, and the resulting signal output was used to determine peaks and nadirs from predetermined time points and a rolling approach. Correlation and residual analysis showed that the Butterworth filter attenuated the “noise” in the signal, while maintaining the integrity of the raw data (r =.9892; mean standardized residual −9.71 × 103 ± 3.80). Means derived from predetermined windows, irrespective of length and data smoothing, underestimated the magnitude of peak and nadir [HHb] compared to a rolling mean approach. Consequently, sprint-induced metabolic changes (inferred from $Δ$[HHb]) were underestimated. Based on these results, we suggest using a digital filter to smooth NIRS data, rather than an arithmetic mean, and a rolling approach to determine peaks and nadirs for accurate interpretation of muscle oxygenation trends.