Purpose This study evaluated prognostic factors for the nonoperative treatment of stiff shoulder. Methods Between June 2005 and May 2010, 497 stiff shoulders treated at our institute were included in this study. Multivariable analysis for recovery with Cox proportional hazard model was performed. The chief determining variable was pathogenesis (idiopathic, diabetic, post-traumatic) and confounding variables were age (49 or less, 50-59, 60 and above), sex, onset to visit interval (three months or less, four months or more), and external rotation (under 0°, 0°or more) or forward flexion (less than 90°, 90°or more) or internal rotation on the first visit. Results There were 356 idiopathic, 61 diabetic, and 80 posttraumatic stiff shoulders. Hazard ratio (HR) and 95 % confidence interval (CI) for recovery (lower HR means poor prognosis) was 0.54 (0.36-0.96) in the diabetic group (p=0.007), and 0.92 (0.67-1.25) in the post-traumatic group (p=0.58) compared with the idiopathic group. A positive correlation was observed in ages of 60 or over (HR 1.46, 95 % CI 0.86-1.65, p-value 0.02) and external rotation under 0°on the first visit (0.71, 0.53-0.96, 0.03). No correlations were observed in sex (p=0.78) or onset to visit interval (p=0.99). Similar results were obtained when forward flexion or internal rotation was used as a confounding variable. Conclusions Diabetes mellitus and severely restricted joint motion on the first visit were poor prognostic factors and ages of 60 or over was a better prognostic factor.