The use of appropriate reference genes is essential for the generation of accurate and biologically meaningful results from quantitative real-time PCR (qRT-PCR) analysis. However, studies have found that the expression of most commonly used reference genes is not always independent of the tissues, treatments, or developmental stages studied. geNormPlus, NormFinder, and BestKeeper, were applied and the expression stability of nine candidate genes was evaluated in different data sets during wheat grain development. Varying degrees of diversity in either single or multiple reference genes were observed among the results generated from the different computer programs, parameters, and data sets. Therefore, the reliability of identified reference genes in the flag leaf and the complete set of samples was estimated by monitoring the expression dynamics of three NAM genes (TaNAM-A1, TaNAM-B1, and TaNAM-B2). The results suggest that a single control gene identified by geNormPlus for use with the complete set of samples, and multiple reference genes selected by geNormPlus and NormFinder exclusively for the flag leaf outperformed others owing to the consistent results with previous analyses of these genes, which were normalized against a verified single control gene. Given the limit of NormFinder in gene numbers of multiple reference genes, robust quantification can be achieved by normalizing against Ta27922 or multiple reference genes chosen by geNormPlus for individual tissues.