Populus euphratica is the only arboreal species that is established in the world's largest shifting-sand desert in China and is well-adapted to the extreme desert environment, so it is widely considered a model system for researching into abiotic stress resistance of woody plants. However, few P. euphratica reference genes (RGs) have been identified for quantitative real-time polymerase chain reaction (qRT-PCR) until now. Validation of suitable RGs is essential for gene expression normalization research. In this study, we screened 16 endogenous candidate RGs in P. euphratica leaves in six abiotic stress treatments, including abscisic acid (ABA), cold, dehydration, drought, short-duration salt (SS) and long-duration salt (LS) treatments, each with 6 treatment gradients. After calculation of PCR efficiencies, three different software tools, NormFinder, geNorm and BestKeeper, were employed to analyze the qRT-PCR data systematically, and the outputs were merged by means of a non-weighted unsupervised rank aggregation method. The genes selected as optimal for gene expression analysis of the six treatments were RPL17 (ribosomal protein L17) in ABA, EF1α (elongation factor-1 alpha) in cold, HIS (histone superfamily protein H3) in dehydration, GIIα in drought and SS, and TUB (tubulin) in LS. The expression of 60S (the 60S ribosomal protein) varied the least during all treatments. To illustrate the suitability of these RGs, the relative quantifications of three stress-inducible genes, PePYL1, PeSCOF-1 and PeSCL7 were investigated with different RGs. The results, calculated using qBasePlus software, showed that compared with the least-appropriate RGs, the expression profiles normalized by the recommended RGs were closer to expectations. Our study provided an important RG application guideline for P. euphratica gene expression characterization.