Learning a new motor task is characterized by improvements in movement smoothness and kinematics repeatability (Shmuelof et al., 2012). Hence, demonstrating biomechanical changes during the early phase of learning can be difficult due to high standard deviations that directly affect statistical analysis. Especially for curve analysis, timing variability results in wide corridor of normality. In addition, in the early phase of learning, each trial participates to the learning process, compromising the collection of redundant data. Besides, for further biomechanical analysis, curves selection/exclusion is often required, and is generally done manually based on a single parameter. The aim of this study was to investigate the interest of cluster analysis (i) to describe biomechanical changes during the early phase of learning manual wheelchair (MWC) locomotion and (ii) to define representative time courses of the locomotion parameters at different steps of the leaning process.