In the present study, we developed new turn scales based on the occurrence of amino acids at each of the four positions of a turn using a structural database comprised of 87 proteins. We found that the scales correctly predicted a fraction of the turn regions in proteins with approximately 80% confidence. We used the turn scales for predicting the location of antigenic sites in proteins. The method was developed with the specific aim of predicting only a few peaks for each protein (two or three). We found that it leads to a high level of accurate prediction (70% of correct prediction of known epitopes). Our method should be useful for selecting protein regions to be synthesized in order to produce anti-peptide antibodies cross-reacting with the parent protein.