Electromechanically assisted gait training is a promising task-oriented approach for gait restoration, especially for people with subacute stroke. However, few guidelines are available for selecting the parameter values of the electromechanical Gait Trainer (GT) (Reha-Stim; Berlin, Germany) and none is tailored to a patient's motor capacity. We assessed 342 GT sessions performed by 20 people with stroke who were stratified by Functional Ambulatory Category. In the first GT session of all patients, the body-weight support (BWS) required was higher than that reported in the literature. In further sessions, we noted a slow reduction of BWS and a fast increment of walking speed for the most-affected patients. Inverse trends were observed for the less-affected patients. In all the patients, the heart rate increment was about 20 beats per minute, even for sessions in which the number of strides performed was up to 500. In addition, the effective BWS measured during GT sessions was different from that initially selected by the physiotherapist. This difference depended mainly on the position of the GT platforms during selection. Finally, harness acceleration in the anteroposterior direction proved to be higher in patients with stroke than in nondisabled subjects. Our findings are an initial step toward scientifically selecting parameters in electromechanically assisted gait training.