Deep brain stimulation is effective at improving motor symptoms of Parkinson's disease. However, the mechanism of action remains unclear and more efficient approaches to stimulation may improve patient quality of life. Here we review how computational models have been used to understand and advance the therapy. We describe two classes of models: (1) abstract models, which aim to replicate behaviors without simulating exact patient measures, and (2) clinically predictive models, which aim to simulate patient specific parameters. Abstract models can be used to develop novel patterns of stimulation while clinically predictive models can be used to aid clinicians in selecting therapeutic stimulation parameters for each patient. These principles can likely be applied to stimulation therapies for a number of disorders.