For development of effective vaccines against viruses, it is of importance to choose appropriate epitopes as the target for immunization. These epitopes should eventually be determined experimentally, but it would be helpful if we could predict candidate epitopes computationally because it accelerates the entire process. To predict candidate epitopes for immunization, it is of great interest to characterize the target epitopes of poliovirus vaccine, which has empirically proven to be the most effective among all vaccines available. Here I show that almost all amino acid sites of poliovirus surface proteins VP1, VP2, and VP3 including neutralization epitopes are negatively selected and no site is under positive selection. These results, together with those obtained in previous studies, indicate that vaccines directed against epitopes, which consist of negatively selected sites protect vaccinees more effectively than those directed against epitopes which contain positively selected sites. These observations suggest that candidate epitopes for immunization are predicted by the molecular evolutionary analysis of viral protein (and its coding nucleotide) sequences, as the epitopes which consist exclusively of negatively selected amino acid sites.