In silico antibody-antigen binding predictions are generally employed in research to rationalize epitope development. These techniques are widely spread despite their technical limitations. To validate the results of these bioinformatic calculations evidence based comparative in vitro studies are necessary. We have used a well-conserved mitochondrial inner membrane antigen-citrate synthase to develop a model for comparative analysis of the predicted and the immunoserologically verified epitopes of circulating autoantibodies. Epitopes were predicted using accepted tools: the GCG Wisconsin package and TEPITOPE 2000. An overlapping multipin ELISA assay--covering 49% of the citrate synthase molecule--was developed to map autoantibody epitopes of individuals (healthy, systemic autoimmune, and heart transplanted) in different immunopathological conditions. From the 40 synthesized decapeptides 34 were predicted in silico and 27 were validated in vitro. Thirty-two percent of epitopes were recognized by majority of sera 47% by at least one sera. False positive predictions were 21%. There was major difference in the recognized epitope pattern under different immunopathological conditions. Our results suggest that special databases are needed for training and weighing prediction methods by clinically well-characterized samples, due to the differences in the immune response under different health status. The development of these special algorithms needs a new approach. A high number of samples under these special immunological conditions are to be mapped and then used for the "fine tuning" of different prediction algorithms.