An experimental proficiency testing survey for serum alpha-fetoprotein was conducted among 16 cooperating laboratories. Samples of placental cord serum serially diluted in normal adult serum (simulated maternal serum) were assayed. Results were analyzed for qualitative and quantitative relative accuracy, precision on duplicate and nonduplicate (dilution-related) samples, parallelism, and comparability of units. Although some results had wide distribution, relative accuracy improved when results were normalized against a standard. This procedure improved the overall geometric standard deviation among laboratories on quantitative results from 1.300 to 1.088 and compressed the limits for consensus ranges of results to 40% of that of the unnormalized data. No significant differences in actual mean values were noted between international units per milliliter and nanograms per milliliter in measuring alpha-fetoprotein concentrations at the precision obtained in this study. In interpreting whether a sample was "low or normal" or "abnormally high" for alpha-fetoprotein, we observed that an 80% or greater consensus was achieved on only 6 of the 10 survey samples because borderline samples were included in the survey.