Serum microRNAs (miRNAs) have become a highlighted research hotspot, especially for their great potential as a novel promising non-invasive biomarker in cancer diagnosis. The most frequently used approach for serum miRNAs detection is quantitative real time polymerase chain reaction (qPCR). In order to obtain reliable qPCR data of miRNAs expression, the use of reference genes as endogenous control is undoubtly necessary. However, no systematic evaluation and validation of reference genes for normalizing qPCR analysis of serum miRNAs has been reported in colorectal adenocarcinoma. We firstly profiled pooled serum of colorectal adenocarcinoma, colorectal adenoma and healthy controls and selected a list of 13 miRNAs as candidate reference genes. U6 snRNA (U6) and above-mentioned 13 miRNAs were included in further confirmation by qPCR. As a result, 5 miRNAs (miR-151a-3p, miR-4446-3p, miR-221-3p, miR-93-5p and miR-3184-3p) were not detected in all samples and 2 miRNAs (miR-197-3p and miR-26a-5p) were relatively low with median Cq more than 35, and were excluded from further stability analysis. Then variable stability of other 6 miRNAs (miR-103b, miR-484, miR-16-5p, miR-3615, miR-18a-3p and miR-191-5p) and U6 were evaluated using two algorithms: geNorm and NormFinder which both identified miR-191-5p as the most stably expressed reference gene and selected miR-191-5p and U6 as the most stable pair of reference genes. After validating in an independent large cohorts and selecting miR-92a-3p as target miRNA to evaluate the effect of reference gene, we propose that combination of miR-191-5p and U6 could be used as reference genes for serum microRNAs qPCR data in colorectal adenocarcinoma, colorectal adenoma and healthy controls.