To determine the reliability of endomyocardial biopsies in the detection of rejection, we took biopsy specimens from 22 autopsied human cardiac allografts by direct visualization. Five specimens were taken from each of four sites: left ventricular free wall, left ventricular septum, right ventricular free wall, and right ventricular septum. The findings in individual biopsy specimens and the summed diagnosis in the five biopsy specimens from each site were graded for rejection by a grading system similar to the Billingham criteria. Grading of the five specimens from each site also was done using criteria recommended by the International Society for Heart and Lung Transplantation. These findings were compared with the rejection grade determined by examination of large tissue sections from the autopsied hearts, used as the standard. Overall, X-Y correlation for single-specimen grading had an r value of 0.792. Grading based on five biopsy specimens from all sites using the modified Billingham criteria grading scheme had an r value of 0.845; the r value was 0.857 for the new grading system. Specificity of findings was high for all grades. Sensitivity of findings, however, was dramatically less for grades 1, 2, and 3. Predictive values for these grades also were less for grades 1, 2, and 3, but to a lesser degree than sensitivity. No difference in detection of rejection was found from specimens taken from different areas of the heart. We conclude that a significant number of hearts with middle grades of rejection may be underestimated by routine biopsies. The criteria for grading of the International Society for Heart and Lung Transplantation had a slight advantage over a more traditional grading system.