Real-time quantitative PCR (RT-qPCR) is a method with high sensitivity and convenience that has been extensively used to analyze the expression level of target genes. A reference gene with a highly stable expression is required to ensure the accuracy of experimental results. However, the report on appropriate reference genes in arrowheads (Sagittaria trifolia) is still limited. In this study, eight candidate reference genes (ACT5, UBQ, GAPDH, CYP, NAC, IDH, SLEEPER and PLA) were selected. The candidate genes were employed in a RT-qPCR assay in different tissues at different developmental stages of the same tissue (including corm, leaf and leafstalk) in arrowheads. Five statistical algorithms, GeNorm, NormFinder, BestKeeper, delta cycle threshold (ΔCt) and RefFinder, were used to evaluate the stability of these genes’ expressions in order to identify the appropriate reference genes. The results showed that UBQ was the optimum reference gene in leaf, leafstalk, root, stolon and corm, IDH exhibited the most stable expression during the expansion of corm, UBQ and PLA were the most stable reference genes in developmental stages of leaf and leafstalk, respectively. Finally, the reliability of reference genes was further confirmed by the normalization of PDS and EXP1 genes under different arrowhead tissues and developmental stages of corm, respectively. This study constitutes important guidance for the selection of reliable reference genes for analyzing the tissue- and developmental-stage-specific expression of genes in arrowheads.