We have evaluated Roche Diagnostics' RIA-CEA and Abbott Diagnostics' EIA-CEA methods for precision, normal reference interval, concordance, and correlation of malignant disease with increase in carcinoembryonic antigen (CEA) in plasma. In examining concordance, we used data on 138 patients with primary carcinomas of the breast, colon, lung, or pancreas, each further classified by extent of dissemination. We find the two methods to be comparably precise. The respective upper reference limits of normal for the Roche and Abbott methods were determined to be 5.0 micrograms/L and 4.6 micrograms/L. The regression equation for a log transformation of the 177 data points is y = 0.966x + 0.03, where x = Roche and y = Abbott, with a correlation coefficient of 0.948. According to the criteria we used, the concordance was 78.7%. The largest discordance was observed in widely disseminated disease states and in cancers of the colon and pancreas. Paired data analysis of discordance indicated greater increases in apparent CEA by the Abbott method in most colon cancers with liver involvement; six of the eight discordant pancreatic cancers had higher Roche-CEA values. CEA heterogeneity and the role of the liver in CEA metabolism appear to contribute to the observed differences. We show why the two methods should not be used interchangeably, and that baseline values for CEA must be established for each method.