Levels of alpha-fetoprotein were determined in 297 paired samples of maternal serum and amniotic fluid from patients with normal pregnancies at 16 to 17 completed weeks of gestation. Gestational ages were determined by menstrual history and ultrasonography, and the assays for alpha-fetoprotein were performed by a single reference laboratory. The results of the study failed to demonstrate any statistical relationship between the concentrations of alpha-fetoprotein in these two fluid compartments (r = 0.006). This finding was confirmed when the results at 16 and 17 weeks were analyzed individually. These findings indicate that amniotic fluid levels of alpha-fetoprotein cannot be predicted by those levels determined in maternal serum, and that simple diffusion is an inadequate explanation for the transfer of alpha-fetoprotein from amniotic fluid to maternal serum. The lack of correlation also emphasizes that the sensitivity and specificity of maternal serum alpha-fetoprotein screening for neural tube defects must continue to be based upon the detection rates in large population studies and the determination of "normal" values by reference laboratories.