This prospective study was undertaken to compare the ability of clinical estimation, diagnostic ultrasonography, magnetic resonance imaging, and arthroscopy to estimate the size of rotator cuff tears. Estimates of rotator cuff tear size were compared with the findings at open operation in 33 consecutive patients with a presumptive diagnosis of rotator cuff tear. Arthroscopy estimates of rotator cuff tear size correlated best with actual tear size (Pearson correlation coefficient r = 0.92; P <.001). Magnetic resonance imaging (r = 0.74; P <.001) was similar to ultrasonography (r = 0.73; P <.001). Estimates of rotator cuff tear size after clinical assessment alone had weaker correlation coefficients (r = 0.41; P =.02) than the other methods. Each method underestimated rotator cuff tear size by 12%, 30%, 33%, and 38%, respectively. No method was able to determine the size of partial-thickness rotator cuff tears (r < 0.02).