Most autoimmune diabetes occurs in those without a diabetic relative, but few cases are identifiable prospectively. To model general population prediction, 491 consecutive newly diabetic children from all of Sweden were tested for autoantibodies to glutamate decarboxylase (GAD65ab), insulin (IAA), and islet cells (ICA), and for HLA-DQ genotypes by PCR; 415 matched control children were tested in parallel. GAD65ab sensitivity/specificity was 70/96%, versus 84/96% for ICA, 56/97% for IAA, 93/93% (any positive), 39/99.7% (all positive), and 41/99.7% (GAD65ab plus IAA). The latter's 25% predictive value was not improved by requiring concomitant high-risk HLA genotypes. GAD65ab were associated with DQA1*0501/B1*0201 (DQ2; P = 0.007) but not DQA1*0301/B1*0302 (DQ8), and IAA with DQA1*0301/B1*0302 (DQ8; P = 0.03) but not DQA1*0501/B1*0201 (DQ2). GAD65ab were more prevalent in females than males (79 vs. 63%; P < 0.0001) but did not vary with onset age nor season. Combining the three antibody assays yielded sufficient sensitivity for screening. GADab were relatively sensitive/specific for diabetes, but even with HLA marker combinations yielded predictive values insufficient for early immunointervention in the low-prevalence general population.