Conventional low-pass filtering of 3D motion capture signals prior to estimating knee joint moments and ACL injury risk has known limitations. This study aimed to evaluate the fractional Fourier filter (FrFF), which employs a time-varying cut-off frequency, for assessing peak knee moments during common ACL injury risk screening tasks. Ground reaction force and motion data were collected from 23 team sport athletes performing 45° unanticipated sidesteps and drop jumps. Peak knee abduction, internal rotation and non-sagittal moments were estimated using inverse dynamics after five different low-pass filter approaches were applied (FrFF vs. four variations of a fourth-order Butterworth filter). The FrFF produced peak knee moments larger than "matched" (i.e. force and motion cut-off frequencies were equivalent) and closer to "unmatched" (i.e. force and motion cut-offs were different) Butterworth filter approaches and removed problems with representing foot-to-ground impact peaks. Participants with larger peak moments were identified as "at risk" of injury irrespective of filter approach, but the FrFF identified "at risk" classifications conventional approaches did not. Preliminary evidence suggests that the FrFF displays enhanced sensitivity to movement strategies that induce high knee loads. This was most evident for sidestepping, with more research warranted to optimise the FrFF for drop jumps.