Introduction: Clinical gait analysis is used to assess patients' impairments, improve treatment decision-making and monitor patients’ progress. Motion capture of markers placed on the skin is the gold standard technique for gait analysis. Markers referenced to the underlying bones are used to define a biomechanical model, i.e. an anatomical frame for each body segment. Several biomechanical models have been proposed, and none can be considered as ground truth. The outcomes from different biomechanical models provide non comparable data, this hampers data sharing and impedes the full potential of clinical interpretation and treatment recommendation. Objectives: This study explores the differences in biomechanical models’ definitions used for gait analysis and evaluates four harmonisation techniques with the aim to improve kinematic data comparability. Methods: Nine healthy participants performed a walking task. A merged markerset was developed by including three widely adopted biomechanical models: CGM, CAST and LAMB. The harmonisation processes involved three coordinate transformations between anatomical frames based on the calibration trial, the mean of the task trial and the frame-by-frame of the task trial. The fourth harmonisation consisted in inverse kinematics fitting based on the same underlying model (OpenSim). Differences before and after each harmonisation approach were analysed by considering joint kinematics. Results: The differences between the native biomechanical models' definitions reach 23.2° of rotation and 41mm of translation. A systematic difference between models was found, which was higher than between subjects, and varied across the gait cycle with a nonlinear pattern. All harmonisation processes improved kinematic data comparability. No statistically significant difference between the curves was found when results were harmonised with frame-by-frame coordinate transformation. Harmonisation based on inverse kinematics provides comparable results, with the exception of ankle parameters on sagittal plane. Discussion: The nonlinear and gait cycle-dependent systematic difference between biomechanical models is suggested to be due to the soft tissue artefact. As soft tissue artefact varies across the gait cycle and between subjects, it is reasonable that a frame-by-frame harmonisation provides more comparable results, as the dynamic transformation takes into account the nonlinear soft tissue behavior. However, it remains specific for gait. Conclusion: This study provides promising methodologies for kinematic data harmonisation and allows to easily switch between biomechanical models without gaining consensus on which biomechanical model should be mostly used. However, further investigation of soft tissue artefact effects on biomechanical models’ definitions and consequent joint kinematics is still required to approximate the ground truth.