A triple-bridge, indirect peroxidase-antiperoxidase method for demonstrating carcinoembryonic antigen (CEA) in frozen, ethanol-fixed or formalin-fixed, paraffin-embedded specimens was evaluated. Examination of 359 tissue specimens--234 malignant tumors, 37 benign neoplasms, 41 nonneoplastic diseased tissues, and 47 normal specimens--showed that CEA could usually be demonstrated in a group of cancers. We could detect CEA in carcinomas of the stomach, colon, rectum, pancreas, lung, and cervix. However, malignant tumors of the breast, prostate, kidney, larynx, brain, lymphoreticular system, soft tissues, and skin proved negative for CEA by the immunoperoxidase test. CEA could be detected in ethanol- or formalin-fixed sections. The only nonmalignant specimens showing CEA staining were a few benign tumors, the mucosae of some cases of colitis, and the resection margins of 2 cases of colon cancer; however, these were commonly very weak reactions. Measurement of tumor CEA content by radioimmunoassay revealed two causes for this relative specificity of the immunoperoxidase test for CEA:1) a quantitative difference existed in tissue CEA among the various specimens, and 2) the threshold for CEA staining in malignant specimens was usually above that in nonmalignant specimens. An analysis of the formalin-paraffin-treated sections showed that immunoperoxidase-tested CEA positivity reflected CEA levels in tissue of at least 3.0-5.0 mug/g; this permitted retrospective estimates of minimal tissue CEA concentrations in older histopathologic specimens by the immunoperoxidase reaction method. Formalin-paraffin-treated sections as old as 10 years still had demonstrable CEA. Although tumor CEA concentration correlated well with immunoperoxidase staining for CEA, plasma CEA titer did not necessarily reflect tumor CEA content. CEA positivity in primary and secondary tumors was strongly correlated; it was less strongly correlated with level of tumor differentiation.