NY-ESO-1 has been a major target of many immunotherapy trials because it is expressed by various cancers and is highly immunogenic. In this study, we have identified a novel HLA-B*1801-restricted CD8(+) T cell epitope, NY-ESO-1(88-96) (LEFYLAMPF) and compared its direct- and cross-presentation to that of the reported NY-ESO-1(157-165) epitope restricted to HLA-A*0201. Although both epitopes were readily cross-presented by DCs exposed to various forms of full-length NY-ESO-1 antigen, remarkably NY-ESO-1(88-96) is much more efficiently cross-presented from the soluble form, than NY-ESO-1(157-165). On the other hand, NY-ESO-1(157-165) is efficiently presented by NY-ESO-1-expressing tumor cells and its presentation was not enhanced by IFN-γ treatment, which induced immunoproteasome as demonstrated by Western blots and functionally a decreased presentation of Melan A(26-35); whereas NY-ESO-1(88-96) was very inefficiently presented by the same tumor cell lines, except for one that expressed high level of immunoproteasome. It was only presented when the tumor cells were first IFN-γ treated, followed by infection with recombinant vaccinia virus encoding NY-ESO-1, which dramatically increased NY-ESO-1 expression. These data indicate that the presentation of NY-ESO-1(88-96) is immunoproteasome dependent. Furthermore, a survey was conducted on multiple samples collected from HLA-B18(+) melanoma patients. Surprisingly, all the detectable responses to NY-ESO-1(88-96) from patients, including those who received NY-ESO-1 ISCOMATRIX™ vaccine were induced spontaneously. Taken together, these results imply that some epitopes can be inefficiently presented by tumor cells although the corresponding CD8(+) T cell responses are efficiently primed in vivo by DCs cross-presenting these epitopes. The potential implications for cancer vaccine strategies are further discussed.