When joint kinematics is analyzed using noninvasive stereophotogrammetry, movements of the skin markers relative to the underlying bone are regarded as artefacts (soft tissue artefact (STA)). Recent literature suggests that an appropriate estimation of joint kinematics may be obtained by compensating for only a portion of the STA, but no evidence for this case has been reported, and which portion of the STA should be selected remains an issue. The aim of this study was to fill this gap. A modal approach was used to represent the STA. This resulted in a series of additive components (modes) and in the possibility to select a subset of them. The following STA definitions were used: individual skin marker displacement (MD), marker-cluster geometrical transformation (GT), and skin envelope shape variation (SV). An STA approximation for each of the three definitions was obtained by ordering modes on the basis of their contribution to the total STA energy and truncating the relevant series at 90% of it. A fourth approximation was obtained when the GT definition was used, by selecting the modes that represented the marker-cluster rigid transformation (i.e., three translation and three rotation modes). The different STA approximations were compared using data obtained during the stance phase of running of three volunteers carrying both pin and skin markers. The STA was measured and knee joint kinematics estimated using four skin marker datasets compensated for the above-mentioned STA approximations. Accuracy was assessed by comparing results to the reference kinematics obtained using pin markers. The different approximations resulted in different numbers of modes. For joint angles, the compensation efficiency across the STA approximations based on an energy threshold was almost equivalent. The median root mean square errors (RMSEs) were below 1 deg for flexion/extension and 2 deg for both abduction/adduction and internal/external rotation. For the joint displacements, the compensation efficiency depended on the STA approximation. Median RMSEs for anterior/posterior displacement ranged from 1 to 4 mm using either MD, GT, or SV truncated series. The RMSEs were virtually null when the STA was approximated using only the GT rigid modes. This result, together with the limited number of modes used by this approximation (i.e., three translations and three rotations of the marker-cluster), makes the STA rigid component and a good candidate for designing an STA model to be incorporated in an enhanced bone pose estimator.