In 1984 we initiated a national external quality assessment (EQA) program (supported by the Italian National Research Council, CNR) for the CEA assay; at present, about 200 Italian laboratories are participating in the program. The laboratories assayed the quality control (QC) samples according to their routine procedures and returned the results together with the name of the method/kit they used. The collected results were computer-processed and reports were sent back to the participants. A significant reduction of the CVt (mean between-laboratory agreement) of the CEA assay was observed throughout the EQA survey (from 35% in 1985 to 20-25% in the last cycles). In order to better clarify the differences in variability observed in the first QC cycles against the last ones, we used the ANOVA technique to evaluate the components of variability. The improvement in between-laboratory agreement was mainly due to the reduction of the between-kit component (from 30.5% to 15.2%), rather than to the smaller decrease observed for the within-kit variability (from 18.4% to 14.0%). The results reported for QC samples from different materials showed differences in the between-lab variability and substantial changes of the kit biases, thus suggesting a different specificity of the antibodies used in the various method/kits against different families of CEA molecules. Considerable uncertainty was also encountered in the clinical classification of low pathological samples, which seems mainly due to the variability in cut-off values used by the laboratories for the clinical assessment of the same analytical results. Our data indicate a progressive increase in the reliability of CEA determination during our study and confirm that EQA has improved the reliability of analysis carried out by the participating laboratories, thus stimulating the kit manufacturers to provide more reliable products.