There is an increasing demand for accurately predicting human movement intentions. To be effective, predictions must be performed as early as possible in the preceding step, though precisely how early has been studied relatively little; how and when a person's movement patterns in a transition step deviate from those in the preceding step must be clearly defined. In this study, we collected motion kinematics, kinetics and electromyography data from 9 able-bodied participants during 7 locomotion modes. Twelve types of steps between the 7 locomotion modes were studied, including 5 continuous steps (taking another step in the same locomotion mode) and 7 transitions steps (taking a step from one locomotion mode into another). For each joint degree of freedom, joint angles, angular velocities, moments, and moment rates were compared between continuous steps and transition steps, and the relative timing during the transition step at which these parameters diverged from those of a continuous step, which we refer to as transition starting times, were identified using multiple analyses of variance. Muscle synergies were also extracted for each step, and we studied in which locomotion modes these synergies were common (task-shared) and in which modes they were specific (task-specific). The transition starting times varied among different transitions and joint degrees of freedom. Most transitions started in the swing phase of the transition step. These findings can be applied to determine the critical timing at which a powered assistive device must adapt its control to enable safe and comfortable support to a user.