In order to precisely assess gene expression levels, the suitable internal reference genes must be served to quantify real‐time reverse transcription polymerase chain reaction (RT‐qPCR) data. For armyworm, Mythimna separata, which reference genes are suitable for assessing the level of transcriptional expression of target genes have yet to be explored. In this study, eight common reference genes, including β‐actin (β‐ACT), 18 s ribosomal (18S), 28S ribosomal (28S), glyceraldehyde‐3‐phosphate (GAPDH), elongation fator‐alpha (EF1α), TATA box binding protein (TBP), ribosomal protein L7 (RPL7), and alpha‐tubulin (α‐TUB) that in different developmental stages, tissues and insecticide treatments of M. separata were evaluated. To further explore whether these genes were suitable to serve as endogenous controls, three software‐based approaches (geNorm, BestKeeper, and NormFinder), the delta Ct method, and one web‐based comprehensive tool (RefFinder) were employed to analyze and rank the tested genes. The optimal number of reference genes was determined using the geNorm program, and the suitability of particular reference genes was empirically validated according to normalized HSP70, and MsepCYP321A10 gene expression data. We found that the most suitable reference genes for the different experimental conditions. For developmental stages, 28S/RPL7 were the optimal reference genes, both RPL7/EF1α were suitable for experiments of different tissues, whereas for insecticide treatments, 28S/α‐TUB were suitable for normalizations of expression data. In addition, 28S/α‐TUB were the suitable reference genes because they have the most stable expression among different developmental stages, tissues and insecticide treatments. Our work is the first report on reference gene selection in M. separata, and might serve as a precedent for future gene expression studies.