Objectives: To assess interobserver reproducibility (agreement and reliability) of visually estimated shoulder range of motion (ROM) and to study the influence of clinical characteristics on the reproducibility. Design: Test-retest analyses. Setting: Various health care settings in the Netherlands. Participants: Consecutive patients with shoulder complaints (N=201) referred by 20 general practitioners, 2 orthopedic physicians, and 20 rheumatologists. Interventions: Not applicable. Main Outcome Measures: Independent visual estimation by 2 physiotherapists of the ROM. Agreement was calculated as the mean difference in visual estimation between examiners ±1.96 × standard deviations of this mean difference. The intraclass correlation coefficient (ICC) was calculated as a measure of reliability, based on a 2-way random effects analysis of variance. Results: The lowest level of agreement was for visual estimation of active and passive elevation (limits of agreement, -43.4 to 39.8 and -46.7 to 41.5, respectively, for the difference between the affected and contralateral sides), for which the level of agreement was most clearly associated with pain severity and disability. The ability to differentiate between subjects was acceptable for all movements for the difference between the affected and contralateral sides (ICCs, >.70) except for horizontal adduction (ICC=.49). Conclusions: Interobserver agreement was low for the assessment of active and passive elevation, especially for patients with a high pain severity and disability. Except for horizontal adduction, visual estimation seems suitable for distinguishing differences between affected and contralateral ROM between subjects.