Previous studies showed that the yield of immunohistochemistry (IHC) staining on endoscopic biopsies may be as good as on surgically removed tissues. However, we noted that some patients showed inconsistent DNA mismatch repair protein (MMRP) stains between biopsies and surgical specimens. In this study, we aimed to investigate factors which are related to the consistence of MMRP evaluation between two pathologists or between different tissues. Two pathologists were asked to diagnose 4 MMRP, both on endoscopic biopsies and surgical materials, in 51 colorectal cancer (CRC) patients, using a single blind method. The consistence of two specimens and inter-observers' variances across different pathologists were compared respectively and the factors related to this variability were analyzed. Among the 816 paired MMRP, 804 (98.5%) pairs showed concordant IHC stains between biopsies and surgical materials, the agreement was almost perfect for MSH6 and PMS2 (kappa = 0.85, 0.85 separately); 804 (98.5%) pairs showed concordant IHC stains between two pathologists, the inter-observer agreement was almost perfect for MSH6 and PMS2 (kappa =0.85, 0.88 separately). Clinical and pathological characteristics analysis showed that biopsy number and TNM stage were related to the variations. Inter-observers variations should be taken into account during MMRP testing in colorectal cancers. Generous endoscopic biopsies could improve the accuracy of endoscopic biopsy for MMRP detection which can be used histological tool in the evaluation of CRC and a promising new prognostic factor for these patients.