Methodological problems with the assay of fructosamine in serum--standardization, matrix effects, and dependence on buffer pH--have been minimized with a method involving colorimetric assay of each specimen and subsequent re-assay after standard addition of 1-deoxy-1-morpholinofructose. Absorbance at optimum wavelength of 540 nm varies linearly with fructosamine concentration to at least 5.5 mmol/L, and between-run precision is about 6% for both patients' specimens and quality control materials. Correction of fructosamine to serum albumin of 40 g/L minimizes the effect of albumin while maintaining transferability of data and reference values. From data on biological variation, the analytical goal for precision (CV) is less than or equal to 2.6%. The square root of the ratio of intra- to interindividual variance is low, indicating that fructosamine concentrations have a high index of individuality; thus conventional population-based reference values are of limited use. Although this assay may be useful in monitoring disease, we doubt that it provides a valid screening test for diabetes.