The important problem in analyzing the paired RNA-seq data is that current methods cannot account for the dependence between the paired groups. Moreover, the Poisson or Negative Binomial distributions used in most methods may not fit the data with biological replicates as assumed. Thus, a novel idea of RNA-seq study is in need to help capture new data characteristics. By modeling the change between two groups with the homogenous birth and death process, we propose a new method (a birth and death model) to detect differentially expressed genes. To make our model more appropriate for the real data, we generalize the original model to a mixed birth and death model. The usefulness and advantages of our approaches are demonstrated on simulations and three real datasets.