In conventional total knee replacement (TKR) surgery, a significant fraction of implants have varus/valgus alignment errors large enough to reduce the lifespan of the implant, so we are developing a more accurate computerassisted procedure aimed at reducing the standard deviation (SD) of the implant procedure. In this study we introduce a new method of locating the ankle joint centre (AJC) using a biaxial model (BM), and determine the accuracy and repeatability of this protocol compared to a digitization method and a spherefitting method used in a current computer-assisted procedure. Repeated in vivo measurements performed by a single operator were obtained from five normal subjects (450 measurements) using the three methods of AJC location. Based on these experiments we estimate the varus/valgus SD of defining the tibial mechanical axis in the frontal plane for the tested population to be 0.28° for the spherical model, 0.17° for the biaxial model, and 0.11° for the conventional digitizing point probe. The mean joint centre locations found by the motionbased models are significantly medial and anterior to the point probe centre.