Abstract Background Health care providers need prognostic factors to distinguish between patients who are likely to recover and those who are not likely to recover. Objective The aim of this study was to: (1) describe the clinical course of recovery and (2) identify prognostic factors of recovery in patients with shoulder pain at the 26-week follow-up. Design A prospective cohort study was carried out in the Netherlands and included 389 patients who consulted a physical therapist for a new episode of shoulder pain. Method Participants were followed for 26 weeks. Potential predictors of recovery were selected from the literature and, with the addition of 2 new variables (ie, use of diagnostic ultrasound and working alliance), evaluated in the multivariable regression analysis. Multiple imputation was used to handle missing data, and bootstrap methods were used for internal validation. Results The recovery rate was 60% for the total population and 65% for the working population after 26 weeks. Short duration of complaints, lower disability scores, having a paid job, better working alliance, and no feelings of anxiety or depression were associated with recovery. In the working population, only duration of complaints and disability remained in the final model. The area under the receiver operating characteristic curve (AUC) for the final model was 0.67 for the total population and 0.63 for the working population. After internal validation, the AUC was corrected to 0.66 and 0.63, respectively. Limitations External validation of the prognostic model should be done prior to its use in clinical practice. Conclusion The results of this study indicate that several factors can predict recovery.