Circulating microRNA (miRNA) expression profiles have been reported to be promising biomarkers for drug-induced liver injury in preclinical and clinical practice. Proper normalization is critical for accurate miRNAs expression analysis. Herein, using SYBR green quantitative real-time PCR (RT-qPCR), we evaluated the expression stability of six candidate reference genes including two commonly used small RNAs (U6, 5S) and four miRNAs (let-7a, miR-92a, miR-103 and miR-16) in plasma of rats with acetaminophen-induced hepatotoxicity. Data were analysed using geNorm, Normfinder, BestKeeper and comparative delta-Ct statistical models, and the results consistently show that miR-103 is the most stably expressed reference gene. Whereas the commonly used housekeeping genes 5S or U6 are all not suitable normalizers, because 5S exhibits extensive variability in expression and U6 has a low expression level across the plasma samples. Then the effect of reference genes on normalization of plasma miR-122 was assessed; when normalized to the most stable reference gene there were significant differences between the acetaminophen-treated group and the vehicle group. However, when the data were normalized to a less stably expressed gene, miR-16, a biased result was obtained. Therefore, we recommend that miR-103 as suitable reference gene for plasma miRNAs analysis for acetaminophen-induced liver injury. Data presented in this paper are crucial to successful biomarker discovery and validation for the diagnosis of the early stage of acetaminophen hepatotoxicity. Copyright © 2013 John Wiley & Sons, Ltd.