Background. In a randomized control trial conducted in patients with Parkinson’s disease, a treadmill training program combined with virtual reality that targeted motor and cognitive aspects of safe ambulation led to fewer falls, compared with treadmill training alone. Objective. To investigate if the 2 types of training differentially affected prefrontal activation and if this might explain differences in fall rates after the intervention. Methods. Sixty-four patients with Parkinson’s disease were randomized into the treadmill training arm (n = 34, mean age 73.1 ± 1.1 years, 64% men, disease duration 9.7 ± 1.0 years) or treadmill training with virtual reality arm (n = 30, mean age 70.1 ± 1.3 years, 71% men, disease duration 8.9 ± 1.1 years). Prefrontal activation during usual, dual-task, and obstacle negotiation walking was assessed before and after 6 weeks of training, using a functional near-infrared spectroscopy system. Results. Treadmill training with and without virtual reality reduced prefrontal activation during walking (P <.001), with specific interactions related to training arm (P =.01), lateralization (P =.05), and walking condition (P =.001). For example, among the subjects who trained with treadmill training alone, prefrontal activation during dual-task walking and obstacle negotiation increased after training, while in the combined training arm, activation decreased. Conclusions. Prefrontal activation during usual and during more challenging walking conditions can be altered in response to 2 different types of training. The addition of a cognitive training component to a treadmill exercise program apparently modifies the effects of the training on the magnitude and lateralization of prefrontal activation and on falls, extending the understanding of the plasticity of the brain in PD.