The mud crab is widely cultured in the coastal areas of southern China because of its short growth cycle, high nutrition value, large market demand, and high economic value. At present, most of the mud crab larvae come from the natural populations, which puts great pressure on its natural resources. Therefore, establishing industry-scale artificial reproduction and using artificial larvae will effectively reduce the harvest of wild resources and protect the natural resources of the mud crab. On the other hand, embryonic development is the basis of artificial reproduction and individual development. At present, the research on embryo development of mud crabs is mainly focused on morphological observation and measurement of physiological and biochemical indexes. Knowledge concerning the mechanisms regulating embryo development at molecular level is poor. Commonly, quantitative real time PCR (qRT-PCR) is used to accurately quantify mRNA expression levels. However, this technique is highly dependent on stably expressed housekeeping genes to avoid experimental errors and variations. Furthermore, the universal ideal reference gene suitable for all experiments has not been obtained. To identify the most stable internal reference genes for studying gene expression in different embryonic developmental stages of mud crabs, tubulin alpha 1a (), ribosomal protein S6 kinase (), glyceraldehyde-3-phosphate dehydrogenase (), 28S ribosomal RNA (28S), 18S ribosomal RNA (18S), and elongation factor 1A () were employed to quantify their expression by qRT-PCR. The qRT-PCR results were used to evaluate their expression stabilities by three algorithms-geNorm, NormFinder, and BestKeeper. The geNorm program analysis showed that the expression of had the highest stability. The results also suggested that it was better to use these two internal reference genes together to correct the quantitative results of target genes. NormFinder analysis showed that had the most stable expression among the nine internal candidate reference genes. BestKeeper program analysis showed that the standard deviation and coefficient of variation of were the lowest among the nine candidate genes. Combining the three internal reference gene screening softwares, , could be used to correct the expression of target genes in different embryonic developmental stages. Furthermore, we found that the expression of many published internal reference genes was not stable enough under our conditions. In our experiment, commonly used internal reference genes such as -actin, 18S, and 28S were not suitable for standardizing gene expression data in the different embryonic developmental stages of .