Context Commercially, velvet antlers along the longitudinal axis are divided into four portions, namely, wax-like (WL), blood-colour (BC), honeycomb-like (HL) and bone (B) slices from the top to the base. However, there is no evidence at a molecular level showing the accuracy of this classification. Aims The aim of the present study was to take transcriptional approach to assess the accuracy of the traditional classification for these four portions of velvet antler, and to link the expressed mRNAs of each portion with possible functions by using bioinformatics analysis. Methods Three sticks of three-branched velvet antlers of sika deer were harvested from three anaesthetised 4-year-old sika deer. On the basis of the traditional methods used commercially, the velvet antler sticks were divided into the four portions of WL, BC, HL and B. Transcriptome sequencing was performed using Illumina HiSeq × Ten at BGI (Shenzheng, China). Key results In total, 5647 genes were obtained from the four portions. Spearman correlation analysis grouped these four portions into three clusters (WL, BC, HL+B). C-means analysis further confirmed a similar trend, indicating the accuracy of the new classification based on transcriptome analysis. Further functional analysis showed that highly expressed genes in WL, BC and HL+B were mainly related to cell cycle, cartilage development, and bone development respectively. Conclusions Four-portion classification based on traditional methods should be replaced by three-portion classification based on the mRNA expression levels. Implications We believe that this new classification can contribute to velvet antler industry, providing more accuracy in the use of velvet antlers as pharmaceuticals.