In quantitative gene expression analysis, normalization using a reference gene as an internal control is frequently performed for appropriate interpretation of the results. Efforts have been devoted to exploring superior novel reference genes using microarray transcriptomic data and to evaluating commonly used reference genes by targeting analysis. However, because the number of specifically detectable genes is totally dependent on probe design in the microarray analysis, exploration using microarray data may miss some of the best choices for the reference genes. Recently emerging RNA sequencing (RNA-seq) provides an ideal resource for comprehensive exploration of reference genes since this method is capable of detecting all expressed genes, in principle including even unknown genes. We report the results of a comprehensive exploration of reference genes using public RNA-seq data from plants such as Arabidopsis thaliana (Arabidopsis), Glycine max (soybean), Solanum lycopersicum (tomato) and Oryza sativa (rice). To select reference genes suitable for the broadest experimental conditions possible, candidates were surveyed by the following four steps: (1) evaluation of the basal expression level of each gene in each experiment; (2) evaluation of the expression stability of each gene in each experiment; (3) evaluation of the expression stability of each gene across the experiments; and (4) selection of top-ranked genes, after ranking according to the number of experiments in which the gene was expressed stably. Employing this procedure, 13, 10, 12 and 21 top candidates for reference genes were proposed in Arabidopsis, soybean, tomato and rice, respectively. Microarray expression data confirmed that the expression of the proposed reference genes under broad experimental conditions was more stable than that of commonly used reference genes. These novel reference genes will be useful for analyzing gene expression profiles across experiments carried out under various experimental conditions.