Though goniometric measurements are commonplace in the field of physical medicine, their reliability has not been adequately established. The present study examined the reliability of placing and reading a universal goniometer under controlled conditions. Two registered physical therapists each obtained repeated angular measurements on 22 separate joint positions. Major joints of the upper and lower extremities were chosen, and for each, multiple positions representing different parts of full range of motion were designated. A healthy young adult man was measured by using a rigidly standardized procedure for positioning his joints. Results indicated extremely high inter- and intraobserver reliability coefficients (r = 0.97, r = 0.98, respectively). When percentage agreement (agreement = +/- 5 degrees) reliability was calculated, the interobserver ratio was 64% and the intraobserver reliability ratios were 73% and 77%. Additionally, 95% confidence intervals were constructed around average differences (interobserver differences = therapist 1-therapist 2; intraobserver differences = observation 1-observation 2) collapsed across all joints and positions. This analysis indicated that, regardless of whether difference scores are derived from within or between observers, repeated measurements under controlled conditions can confidently be expected to fall within approximately four angular degrees of each other. Results were compared to those from other studies, limitations of generalizability were discussed, and recommendations for future research were proferred. The authors suggested that the present findings be construed as global upper limit estimates of applied goniometric reliability.