The reliability of databases on the isoflavone composition of foods designed to estimate dietary intakes is contingent on the assumption that soy foods are consistent in their isoflavone content. To validate this, total and individual isoflavone compositions were determined by HPLC for two different soy protein isolates used in the commercial manufacture of soy foods over a 3-year period (n = 30/isolate) and 85 samples of 40 different brands of soy milks. Total isoflavone concentrations differed markedly between the soy protein isolates, varying by 200-300% over 3 years, whereas the protein content varied by only 3%. Total isoflavone content varied by up to 5-fold among different commercial soy milks and was not consistent between repeat purchases. Whole soybean milks had significantly higher isoflavone levels than those made from soy protein isolates (mean +/- SD, 63.6 +/- 21.9 mg/L, n = 43, vs 30.2 +/- 5.8 mg/L, n = 38, respectively, p < 0.0001), although some isolated soy protein-based milks were similar in content to "whole bean" varieties. The ratio of genistein to daidzein isoflavone forms was higher in isolated soy protein-based versus "whole bean" soy milks (2.72 +/- 0.24 vs 1.62 +/- 0.47, respectively, p < 0.0001), and the greatest variability in isoflavone content was observed among brands of whole bean soy milks. These studies illustrate large variability in the isoflavone content of isolated soy proteins used in food manufacture and in commercial soy milks and reinforce the need to accurately determine the isoflavone content of foods used in dietary intervention studies while exposing the limitations of food databases for estimating daily isoflavone intakes.