Smoothing to obtain accurate position and velocity data near impacts in sport biomechanics studies is complicated by the accelerations of impact, endpoint effects and the smoothing technique used. Wrist goniometric data with two levels of random noise added were used to examine three endpoint modelling conditions for obtaining accurate position and velocity data at impact in the tennis forehand. The common approach of smoothing through impact created distortions of the position signal up to 100 ms before impact and resulted in consistent underestimations (-3.2%) of wrist angle and angular velocity (-67.9%) at impact. New linear and polynomial extrapolation conditions smoothed with all techniques provided lower root mean squared errors than the smoothing-through-impact condition, with discrete wrist positions at impact within 1.1% of the criterion data. Both the new linear and polynomial conditions can be used to make more accurate angular position and velocity estimates for tennis impacts, whereas the smoothing-through-impact condition creates spurious decreases in speed before impact that in the past have been assumed to be aspects of skillful movement.