Three-dimensional ankle joint moments were calculated in two separate coordinate systems, from 18 healthy men during the stance phase of walking, and were then compared. The objective was to determine the extent of differences in the calculated moments between these two commonly used systems and their impact on interpretation. Video motion data were obtained using skin surface markers, and ground reaction force data were recorded from a force platform. Moments acting on the foot were calculated about three orthogonal axes, in a global coordinate system (GCS) and also in a segmental coordinate system (SCS). No differences were found for the sagittal moments. However, compared to the SCS, the GCS significantly (p < .001) overestimated the predominant invertor moment at midstance and until after heel rise. It also significantly (p < .05) underestimated the late stance evertor moment. This frontal plane discrepancy was attributed to sensitivity of the GCS to the degree of abduction of the foot. For the transverse plane, the abductor moment peaked earlier (p < .01) and was relatively smaller (p < .01) in the GCS. Variability in the transverse plane was greater for the SCS, and attributed to its sensitivity to the degree of rearfoot inversion. We conclude that the two coordinate systems result in different calculations of nonsagittal moments at the ankle joint during walking. We propose that the body-based SCS provides a more meaningful interpretation of function than the GCS and would be the preferred method in clinical research, for example where there is marked abduction of the foot.