Many current RNA-sequencing data analysis methods compare expressions one gene at a time, taking little consideration of the correlations among genes. In this study, we propose a method to convert such an one-dimensional comparison approach into a two-dimensional evaluation of the ratio of standard deviations (SD) of two constructed random variables. This method allows the identification of differentially expressed genes while controlling a preset significance level conditional on the read count mean-variance relationship. Meanwhile, correlations among genes are naturally accommodated due to the clustering of genes with similar distribution in the proposed σ-σ plot. The proposed distribution-free method is designated as DFseq, because it does not depend on a parametric distribution to fit read count. As a result, compared with parametric methods, DFseq can effectively handle genes with a bimodal-like distribution and/or genes with excessive 0 read counts, as well as genes with outlying observations. Besides, DFseq is an ideal platform for comparing performance of different differential gene expression detection methods.