This experiment tested if balance performance differed between a standardized treadmill surface perturbation task and a clinical pull test and was affected by medication or the presence of body weight support in people with Parkinson's disease (PD). Twenty-seven individuals were tested (14 PD in both ON- and OFF-medication states). Clinical pull test and rapid forward (backward fall) translations of the support surface were applied to induce postural reactions requiring at least 1 step to restore balance. The effects of pull type (clinical vs. treadmill), partial bodyweight support (0 vs 20% body weight) and group (control, PD ON-meds and PD OFF-meds) on reactive stepping as well as practice/learning effect were examined. The number of steps taken and the first step duration were entered in linear repeated-measures mixed-effect models separately. The effects of pull type, group, and bodyweight support were all significant in both metrics, as was ON- vs. OFF-medication. A significant interaction term (group x pull type) was found in the first step duration, showing that the group difference was greater in treadmill compared to the clinical pull test. A significant practice effect was also observed within and across testing sessions. A standardized treadmill perturbation performed slightly better than the classical pull test in distinguishing between groups, and partial weight support did not substantially degrade the test's performance to detect the balance deficits in people with PD.