The large number of mild traumatic brain injuries (mTBIs) occurring each year, along with increasing evidence that these injuries are more serious than previously thought, has created a demand to understand how mTBIs occur so that they can be prevented. The use of helmet-based instrumentation to estimate head impact biomechanics in sports is an increasingly popular method by which to study mTBI in real-world scenarios. One of these helmet-based systems, the HIT System for ice hockey, has limited validation data available. Therefore, the goal of this study was to compare accelerations calculated by this system to reference accelerations measured by an anthropometric test device (ATD). A Hybrid III 50th percentile male ATD head and neck was rigidly mounted and fit with an Easton S9 helmet instrumented with the HIT System for ice hockey. A pneumatic linear impactor was used to impact the helmet at multiple speeds and in different directions with three different interfaces between the ATD head and the helmet:​ a nylon cap, a dry human hair wig, and a human hair wig wet with water to simulate perspiration. Relationships between peak resultant accelerations as measured by the system and the ATD were defined via linear regression, and average errors in helmet-based system measures were calculated. The error was markedly reduced by applying impact direction-specific calibration factors as defined by the regression relationships to the system measurements. The study showed that there is a strong correlation between helmet instrumentation-measured head acceleration for ice hockey and reference acceleration as measured by the ATD. However, the relationship is not one-to-one, it varies by the interface between the ATD head and the helmet, and it varies by impact direction. These findings are important to account for in the analysis of on-ice data collected using this system.