Background: There is a lack of recommendations for selecting the most appropriate gait measures of Parkinson's disease (PD)-specific dual-task costs to use in clinical practice and research. Objective: We aimed to identify measures of dual-task costs of gait and turning that best discriminate performance in people with PD from healthy individuals. We also investigated the relationship between the most discriminative measures of dual-task costs of gait and turning with disease severity and disease duration. Methods: People with mild-to-moderate PD (n=144) and age-matched healthy individuals (n=79) wore 8 inertial sensors while walking under single and dual-task (reciting every other letter of the alphabet) conditions. Outcome measures included 26 objective measures within four gait domains (upper/lower body, turning and variability). The area under the curve (AUC) from the receiver-operator characteristic plot was calculated to compare discriminative ability of dual-task costs on gait across outcome measures. Results: PD-specific, dual-task interference was identified for arm range of motion, foot strike angle, turn velocity and turn duration. Arm range of motion (AUC=0.73) and foot strike angle (AUC=0.68) had the largest AUCs across dual-task costs measures and they were associated with disease severity and/or disease duration. In contrast, the most commonly used dual-task gait measure, gait speed, showed an AUC of only 0.54. Conclusion: Findings suggest that people with PD rely more than healthy individuals on executive-attentional resources to control arm swing, foot strike, and turning, but not gait speed. The dual-task costs of arm range of motion best discriminated people with PD from healthy individuals.