This study characterized the dynamics of islet cell antibodies (ICA), insulin antibodies (IAA), glutamic acid decarboxylase antibodies (GADA), and IA-2 antibodies (IA-2A) in 1006 children recruited from the general population due to human leukocyte antigen (HLA) DQB1-conferred risk for type 1 diabetes (T1D). By the age of 5 yr, 13.8% of the children had had one or more autoantibodies in at least one sample drawn at 3- to 12-month intervals from birth, whereas 6.1% had had one or more of the three autoantibodies to biochemically defined antigens in at least two consecutive samples. The cumulative frequencies of positivity for at least two antibodies ranged from 3.2-4.4%. Seventy-five children (7.5%) had at least once ICA, 83 (8.3%) had IAA, 46 (4.6%) had GADA, and 33 (3.3%) had IA-2A. IAA were transient more frequently than the other antibodies (P < or = 0.03) and fluctuated between positivity and negativity more often than ICA (P = 0.001). The genetically high risk children were positive for each autoantibody reactivity more often (P < or = 0.03) than the moderate risk subjects. Thirteen of the 1006 children (1.3%) presented with T1D by the age of 5 yr. The most sensitive predictors of T1D were ICA and IAA, whereas the most specific predictor was IA-2A. Positivity for at least two autoantibodies of IAA, GADA, and IA-2A had the highest positive predictive value for T1D (34%). We conclude that the frequency of various diabetes-associated autoantibodies increases at a relatively stable rate at least up to the age of 5 yr. Persistent positivity for two or more autoantibodies appears to reflect destructive progressive beta-cell autoimmunity, whereas positivity for a single autoantibody may represent harmless nonprogressive or even regressive beta-cell autoimmunity.