Cross-reactive T cells that recognize both Tep69 (dominant nonobese diabetic (NOD) T cell epitope in ICA69 (islet cell autoantigen of 69 kDa)) and ABBOS (dominant NOD T cell epitope in BSA) are routinely generated during human and NOD mouse prediabetes. Here we analyzed how systemic administration of these mimicry peptides affects progressive autoimmunity in adoptively transferred and cyclophosphamide-accelerated NOD mouse diabetes. These models were chosen to approximate mid to late stage prediabetes, the typical status of probands in human intervention trials. Unexpectedly, high dose (100 microg) i.v. ABBOS prevented, while Tep69 exacerbated, disease in both study models. Peptide effects required cognate recognition of endogenous self-Ag, because both treatments were ineffective in ICA69null NOD congenic mice adoptively transferred with wild-type, diabetic splenocytes. The affinity of ABBOS for NOD I-A(g7) was orders of magnitude higher than that of Tep69. This explained 1) the expansion of the mimicry T cell pool following i.v. Tep69, 2) the long-term unresponsiveness of these cells after i.v. ABBOS, and 3) precipitation of the disease after low dose i.v. ABBOS. Disease precipitation and prevention in mid to late stage prediabetes are thus governed by affinity profiles and doses of therapeutic peptides. ABBOS or ABBOS analogues with even higher MHC affinity may be candidates for experimental intervention strategies in human prediabetes, but the dose translation from NOD mice to humans requires caution.