Physical therapy is an intervention commonly used in the treatment of subjects with frozen shoulder symptoms, with limited proven effect. The purpose of this study was to identify the kinematic features of patients with frozen shoulder who are more likely to respond to physical therapy. Thirty-four subjects presenting frozen shoulder syndrome were studied to determine altered shoulder kinematics and functional disability. Subjects received the same standardized treatment with passive mobilization/stretching techniques, physical modalities (i.e. ultrasound, shortwave diathermy and/or electrotherapy) and active exercises twice a week for 3 months. Initially, subjects were asked to perform full active motion in 3 tests: abduction in the scapular plane, hand-to-neck and hand-to-scapula. During the test, shoulder kinematics were measured using a 3-D electromagnetic motion-capturing system. In the initial and follow-up sessions, the self-reported Flexilevel Scale of Shoulder Function (FLEX-SF) was used to determine functional disability from symptoms. Improvement with treatment was determined using percent change in FLEX-SF scores over three months of treatment [(final score-initial score)/initial score x 100, >20% improvement and < = 20% nonimprovement]. Shoulder kinematics were first analysed for univariate accuracy in predicting improvement and then combined into a multivariate prediction method. A prediction method with two variables (scapular tipping >8.4 degrees during arm elevation, and external rotation >38.9 degrees during hand to neck) were identified. The presence of these two variables (positive likelihood ratio=15.71) increased the probability of improvement with treatment from 41% to 92%. It appears that shoulder kinematics may predict improvement in subjects with frozen shoulder syndrome. Prospective validation of the proposed prediction method is warranted.