Bone marrow mesenchymal stem cells (BMSCs) are considered as multipotent cells, representing a multi-lineage potential to differentiate into mesodermal lineages of mesenchymal tissues, including cartilage, bone, fat, muscle and tendon. Tissue engineering in BMSCs has made great advances in the regeneration of cartilage and bone defects. To uncover the mechanisms of the multipotent differentiation process, the molecular changes in gene expression profiles during chondrogenic and osteogenic differentiation need to be evaluated with reliable, accurate, fast and sensitive methods. Reverse transcription-quantitative polymerase chain reaction is a commonly used technology for analyzing gene expression, depending on an appropriate reference gene to normalize the errors. The commonly used reference genes vary, and no ideal and universal reference genes suitable for all conditions exist; therefore validation of the stability of gene expression is required. In the present study, three common statistical algorithms, geNorm, Normfinder and BestKeeper, were used to identify the expression stability of 12 genes, and the target differentiation markers during the differentiation of BMSCs were evaluated accurately. Our results demonstrated that YWHAZ, PPIA and GAPDH were suitable as reference genes for chondrogenic differentiation, while RPL13a allowed an efficient normalization expression value of interest genes for osteogenic differentiation of BMSCs. By contrast, the most unstable reference genes were 18s rRNA, B2M and HPRT1 in all studies, and these should be avoided when investigating the differentiation of BMSCs. Our results demonstrate validation of the appropriate reference genes for accurate gene expression in chondrogenic and osteogenic differentiation of BMSCs.