The results of 45 MR examinations were prospectively compared with the clinical course and biopsy results in 38 renal transplant patients to determine the role of MR in evaluating allograft dysfunction. Twenty-six patients underwent allograft biopsy. In eight patients in whom the biopsy was performed more than 48 hr after MR examination and in 19 patients who did not have a biopsy, the subsequent clinical course was sufficiently diagnostic to determine the specific cause of the transplant dysfunction. Corticomedullary differentiation, graded from 0 to 3, was not helpful in separating rejection (n = 20) from acute tubular necrosis (n = 9), drug toxicity (n = 7), pyelonephritis (n = 2), or normal grafts (n = 7) because of overlap between groups (sensitivity =; 60%, specificity = 60%). In the six patients with two or more MR studies, serial changes in corticomedullary differentiation were not consistent and could not be used to diagnose rejection. When any abnormality of allograft sinus fat, size or shape, or corticomedullary differentiation was considered, the sensitivity for the diagnosis of rejection approached 80%; however, specificity was low (48%). We conclude that MR imaging is not sufficiently accurate to replace transplant biopsy and therefore has a limited role in the evaluation of transplant dysfunction.