The prevalence of antibodies to the 65 kDa isoform of glutamic acid decarboxylase (GADA) was compared with that of islet cell antibodies (ICA) in 614 non-diabetic Estonian children (314 males) aged 3-18 years representing the general population. GADA were analyzed with a radioligand assay, and ICA with a standard immunofluorescence method with a detection limit of 2.5 Juvenile Diabetes Foundation (JDF) units. Fourteen subjects (2.3%, 95% confidence interval [CI] 1.1-3.5%) tested positive for GADA (median level 10.8 relative units [RU], range 7.7-154.2 RU), while 10 (1.6%, CI 0.6-2.6%) had ICA (median levels 34 JDF units, range 3-97 JDF units). Five subjects (0.8%, CI 0.1-1.5%; p = 0.03 vs GADA and 0.15 vs ICA) were double positive. The individual with the second highest GADA level (129.3 RU) and the highest ICA level (97 JDF units) presented with type 1 diabetes 4 months later. A follow-up sample was obtained approximately 3-4 years after the first sampling in 14 subjects initially positive for ICA and/or GADA. Four of the nine initially ICA-positive children remained positive, but their levels decreased from a median of 42 to 18 JDF units (p = 0.06). Only two of the nine retested subjects initially positive for GADA remained positive in the second sample. These observations suggest that the prevalence of GADA in non-diabetic children is of the same magnitude as that of ICA. Combined positivity for both GADA and ICA is less prevalent than single antibody specificities, indicating that double autoantibody positivity may have a higher predictive value for future type 1 diabetes in the general population than either antibody separately. The evanescent character of diabetes-associated autoantibodies in a proportion of the unaffected children implies that more subjects may experience self-restricted beta-cell damage than the number progressing to actual disease.