Objective. This study aims to extract motor ingredients through data mining from wearable sensors in a virtual reality goal-directed shoulder rehabilitation (GDSR) system and to examine their effects toward clinical assessment. Design. A single-group before/after comparison. Setting. Outpatient research hospital. Subjects. 16 patients with frozen shoulder. Interventions. The rehabilitation treatment involved GDSR exercises, hot pack, and interferential therapy. All patients first received hot pack and interferential therapy on the shoulder joints before engaging in the exercises. The GDSR exercise sessions were 40 minutes twice a week for 4 weeks. Main Measures. Clinical assessments included Constant and Murley score, range of motion of the shoulder, and muscle strength of upper arm as main measures. Motor indices from sensor data and task performance were measured as secondary measures. Results. The pre- and posttest results for task performance, motor indices, and the clinical assessments indicated significant improvement for the majority of the assessed items. Correlation analysis between the task performance and clinical assessments revealed significant correlations among a number of items. Stepwise regression analysis showed that task performance effectively predicted the results of several clinical assessment items. Conclusions. The motor ingredients derived from the wearable sensor and task performance are applicable and adequate to examine and predict clinical improvement after GDSR training.