Diabetes Autoimmunity Study in the Young (DAISY) has followed 1972 children for islet autoimmunity and diabetes: 837 first-degree relatives of persons with type 1 diabetes and 1135 general population newborns identified through human leukocyte antigen (HLA) screening. During follow-up of 4.06 yr (range, 0.17-9 yr), serial determination of autoantibodies to glutamic acid decarboxylase, protein tyrosine phosphatase IA2, and insulin has generated approximately 20,000 results. Among 162 children with at least one positive autoantibody, in 31% the test was false positive (autoantibodies were negative twice on blinded duplicate aliquots), in 31% it was transiently positive (confirmed on blinded duplicate aliquots but negative on follow-up), and in 36% it was persistently positive. Using proportional hazards modeling, the HLA-DR3/4 DQ8 genotype, another positive autoantibody at the first positive visit, and level of autoantibody were predictive of persistent positivity. Only HLA-DR3/4 DQ8 genotype was predictive of progression to diabetes in proportional hazards modeling. This prospective study reveals that cross-sectional determination of islet autoantibodies in a population with relatively low previous probability of autoimmunity identifies as "positive" a large number of individuals who are either false or transiently positive. Predictive value of autoantibodies increases with blinded duplicate and independent sample retesting and incorporation of the level of autoantibody in the predictive algorithm.