Combinatorial libraries of immunoglobulin genes in "phage display" vectors are a powerful tool for obtaining antigen-specific antibody fragments. To date, this approach has been used to isolate abundant, but not rare, human autoantibodies of IgG class. We have compared the relative efficiencies of panning pComb3 libraries made from intrathyroidal plasma cells for abundant human autoantibodies to thyroid peroxidase (TPO) and rare autoantibodies to the thyrotropin receptor (TSHR). TPO-specific Fab were readily obtained from a library using three different forms of recombinant antigen, (i) purified TPO, (ii) impure TPO in culture medium and, (iii) TPO expressed on the surface of CHO cells. In contrast, TSHR-specific Fab were not isolated. This was the case despite repeated pannings of six libraries from three optimal patients (IgG/kappa and IgG/lambda libraries for each patient). Both purified recombinant TSHR and CHO cells expressing TSHR on their surface were used. Library enrichment was observed on some screenings. However, Fab expressed by individual clones or from enriched libraries were not specific as determined by (i) binding to purified, radio-labeled antigen, (ii) FACS analysis of TSHR on intact CHO cells and, (iii) inhibition of radiolabeled TSH binding. Remarkably, in screening for both TPO- and TSHR-specific Fab, neither library enrichment nor the retention of cDNA inserts of the correct size correlated with obtaining Fab with the antigenic specificity sought. Indeed, excellent enrichment could be observed with conditioned medium from untransfected cells. Our data suggest that the key to isolating rare antibodies from phage display libraries is not the creation of vast libraries of greater diversity or even the development of more stable vectors. Rather, success in this endeavor appears to require reducing the "noise" of non-specific clones in a moderately sized library.