Colorectal medullary carcinoma, recognized by the World Health Organization as a distinct histologic subtype, is commonly regarded as a specific entity with an improved prognosis and unique molecular pathogenesis. A fundamental but as yet unaddressed question, however, is whether it can be diagnosed reproducibly. In this study, by analyzing 80 colorectal adenocarcinomas whose dominant growth pattern was solid (thus encompassing medullary carcinoma and its mimics), we provided a detailed description of the morphological spectrum from "classic medullary histology" to nonmedullary poorly differentiated histologies and demonstrated significant overlapping between categories. By assessing a selected subset (n=30) that represented the spectrum of histologies, we showed that the interobserver agreement for diagnosing medullary carcinoma by using 2010 World Health Organization criteria was poor; the κ value among 5 gastrointestinal pathologists was only 0.157 (95% confidence interval, 0.127-0.263; P=.001). When we arbitrarily classified the entire cohort into "classic" and "indeterminate" medullary tumors (group 1, n=19; group 2, n=26, respectively) and nonmedullary poorly differentiated tumors (group 3, n=35), groups 1 and 2 were more likely to exhibit mismatch repair protein deficiency than group 3 (P<.001); however, improved survival could not be detected in either group compared with group 3. Our findings suggest that the diagnosis of medullary carcinoma, as currently applied, may only serve as a morphological descriptor indicating an increased likelihood of mismatch-repair deficiency. Additional evidence including a more objective classification system is needed before medullary carcinoma can be regarded as a distinct entity with prognostic relevance. Until such evidence becomes available, caution should be exercised when making this diagnosis, as well as when comparing results across different studies.