Quantitative real-time reverse transcription polymerase chain reaction (qPCR), a sensitive technique for gene expression analysis, depends on the stability of the reference genes used for data normalization. Caragana intermedia, a native desert shrub with strong drought-resistance, sand-fixing capacity and high forage value that is widespread in the desert land of west and northwest China, has not been investigated regarding the identification of reference genes suitable for the normalization of qPCR data. In this study, 10 candidate reference genes were analyzed in C. intermedia subjected to different abiotic (osmotic, salt, cold and heat) stresses, in two distinct plant organs (roots and leaves). The expression stability of these genes was assessed using geNorm, NormFinder and BestKeeper algorithms. The best-ranked reference genes differed across the different sets of samples, but UNK2, PP2A and SAND were the most stable across all tested samples. UNK2 and SAND would be appropriate for normalizing gene expression data for salt-treated roots, whereas the combination of UNK2, SAND and EF-1α would be appropriate for salt-treated leaves. UNK1, UNK2 and PP2A would be appropriate for PEG-treated (osmotic) roots, whereas the combination of TIP41 and PP2A was the most suitable for PEG-treated leaves. SAND, PP2A and TIP41 exhibited the most stable expression in heat-treated leaves. In cold-treated leaves, SAND and EF-1α were the most stably expressed. To further validate the suitability of the reference genes identified in this study, the expression levels of DREB1 and DREB2 (homologs of AtDREB1 and AtDREB2) were studied in parallel. This study is the first systematic analysis for the selection of superior reference genes for qPCR in C. intermedia under different abiotic stress conditions, and will benefit future studies on gene expression in C. intermedia and other species of the leguminous genus Caragana.