The assessment of the microsatellite instability (MSI) status in colorectal cancers is presently warranted for three reasons: 1) as a screening tool for hereditary nonpolyposis colorectal cancer, 2) as a prognostic marker, and 3) as a potential predictive factor of chemotherapy response. The aim of this study was to evaluate, on a large scale with tissue samples coming from a number of different sources, the difficulties met with routine use of immunohistochemistry (IHC) and to determine if it really does offer an accurate alternative to PCR genotyping. Colorectal carcinomas from 462 consecutive patients resected in public or private hospitals were assessed for MSI status by two methods: MSI testing (with BAT-26 microsatellite) and IHC detection of hMLH1, hMSH2, and hMSH6 proteins. Of the 398 cancers tested, immunohistochemistry was noncontributory in 42 (10.5%), focal in 9 (2.3%), and discordant with the PCR results in 36 (9%). For these 87 cases, complementary analyses were performed to explain discrepancy. After additional IHC assay with modified processing protocols, 8 cases remained noncontributory, 2 focal, and 28 discordant: 18 microsatellite stability IHC/MSI PCR and 10 MSI IHC/microsatellite stability PCR. For these discordant cases, we performed a multiplex PCR assay on DNA extracted from the frozen sample and BAT-26 was amplified from DNA extracted from the paraffin blocks used for IHC. Four discordant cases were reclassified after PCR multiplex assay (3 as MSI and 1 as microsatellite stability). Five other cases displayed intratumoral heterogeneity and 19 remained discordant. The discrepancy could be partly explained by variable technical protocols of fixation in the different laboratories, leading to variations in staining quality and difficulties in IHC interpretation. This population-based study is the first one to show that IHC is not sensitive and specific enough to be used routinely. Immunohistochemistry analysis of MMR proteins must be performed in standardized conditions and interpreted by confirmed pathologists. It cannot replace PCR as long as protocols are not optimized and harmonized.