Abstract: Quantitative real-time PCR (qPCR) is widely used for gene expression analysis, but its accuracy critically depends on stable internal reference genes for normalization. In marine invertebrates, especially non-model taxa such as cephalopods, systematic evaluation of reference genes is limited, leading to potential bias. The cuttlefish Sepiella japonica is ecologically and economically important in China, yet previous molecular studies have often relied on single unvalidated reference genes, which may compromise data reliability. This study aimed to systematically evaluate the stability of five commonly used reference genes (18S, ef-1α, ef-1γ, gapdh, and β-actin) across multiple tissues and sexes of S. japonica, and to identify the most suitable reference genes and optimal number for qPCR normalization. Fifteen to sixteen tissue types were collected from ten healthy adults (five males and five females). Total RNA was extracted, reverse-transcribed, and analyzed by qPCR. Gene stability was assessed using four algorithms (geNorm, NormFinder, BestKeeper, and ΔC_t) integrated with RefFinder, and the optimal gene number was determined using geNorm pairwise variation (V_n/n+1 < 0.15). Four transcriptome-derived genes (creld2, cd109, acy1, and miox) were used for validation. The C_t values of the five genes ranged from 15.47 to 20.83. β-actin and gapdh showed pronounced variability in expression stability among tissues and sexes, indicating their limited suitability for normalization. 18S exhibited the highest expression (mean C_t: 15.47-16.29) and lowest variability but displayed sex-biased expression, whereas ef-1α and ef-1γ remained consistently stable across most tissues in both sexes, with ef-1α being the most robust and showing no sex-related bias. Although specific rankings varied among tissues and sexes, the comprehensive results indicated that ef-1α and ef-1γ possessed the highest overall stability, followed by 18S, while β-actin and gapdh were the least stable. The final comprehensive rankings were ef-1γ > ef-1α > 18S > gapdh > β-actin (male) and ef-1α > ef-1γ > 18S > gapdh > β-actin (female). geNorm analysis (V_2/3 < 0.15) indicated that two genes, mainly ef-1α and ef-1γ, were generally sufficient for reliable normalization in most tissues. Validation confirmed that normalization using the stable ef-1α and ef-1γ accurately reflected the expression differences among tissues, whereas β-actin and gapdh can bias or confound statistical analyses. ef-1α and ef-1γ are identified as the most reliable reference gene combination for qPCR analysis in S. japonica, while 18S can serve as an auxiliary gene for within-sex comparisons. The use of β-actin or gapdh alone is not recommended. This study establishes a systematic framework for selecting reliable reference genes in S. japonica, thereby facilitating robust qPCR normalization and providing a foundation for future gene expression research in S. japonica and other cephalopods.