Diffuse p53 immunostaining distinguishes 85% of serous (Type II) from endometrioid (Type I) carcinomas and is an independent marker for poor prognosis. Interobserver reproducibility for the diagnosis of these entities, as well as selection and prediction of p53 immunostaining results, is unknown. Reproducibility of three pathologists regarding: (1) a two (I and II) and (2) three part classification (I, II or indeterminate); (3) recommendation for p53 staining and (4) expectations of p53 staining results were computed with the kappa (k) statistic. All cases were immunostained for p53 and independently scored. A two and three tiered classification scheme achieved high (k=0.71) and moderate (k=0.49) reproducibility. Non-unanimous cases were more likely to be reclassified into an 'indeterminate' category (27 cases, 39% of passes) compared to those with unanimous (82 cases, 14% of passes) classification. Pathologists recommended p53 immunostaining with poor (k=0.28) reproducibility, but staining prediction was made with good concordance (69%, k=0.50). Moreover, p53 staining was more common in diagnostically discordant (46%) compared to concordant (16%) cases. A subset of endometrial cancers do not readily fit within a two-class system and can be culled from cases that (1) do not achieve interobserver concordance and (2) are more likely to be chosen for p53 immunostaining and (3) are more likely to stain positive for p53. Because p53 is an important marker for endometrial adenocarcinoma outcome, and cannot be predicted in advance in indeterminate cases, p53 immunostaining should be employed in cases with observer disagreement in a binary system.