Accumulating evidence indicates that recognition by TCRs is far more degenerate than formerly presumed. Cross-recognition of microbial Ags by autoreactive T cells is implicated in the development of autoimmunity, and elucidating the recognition nature of TCRs has great significance for revelation of the disease process. A major drawback of currently used means, including positional scanning synthetic combinatorial peptide libraries, to analyze diversity of epitopes recognized by certain TCRs is that the systematic detection of cross-recognized epitopes considering the combinatorial effect of amino acids within the epitope is difficult. We devised a novel method to resolve this issue and used it to analyze cross-recognition profiles of two glutamic acid decarboxylase 65-autoreactive CD4(+) T cell clones, established from type I diabetes patients. We generated a DNA-based randomized epitope library based on the original glutamic acid decarboxylase epitope using class II-associated invariant chain peptide-substituted invariant chains. The epitope library was composed of seven sublibraries, in which three successive residues within the epitope were randomized simultaneously. Analysis of agonistic epitopes indicates that recognition by both TCRs was significantly affected by combinations of amino acids in the antigenic peptide, although the degree of combinatorial effect differed between the two TCRs. Protein database searching based on the TCR recognition profile proved successful in identifying several microbial and self-protein-derived mimicry epitopes. Some of the identified mimicry epitopes were actually produced from recombinant microbial proteins by APCs to stimulate T cell clones. Our data demonstrate the importance of the combinatorial nature of amino acid residues of epitopes in molecular mimicry.