Due to the small amount of starting material and the low capture efficiency of the current technologies, scRNA-seq data contains a large number of zero counts due to the dropout phenomenon. The unique features of scRNA-seq data have led to the development of novel methods for differential expression (DE) analysis. However, few of the existing DE methods for scRNA-seq data estimate the number of molecules pre-dropout and therefore do not explicitly distinguish technical and biological zeroes. We develop DECENT, a DE method for scRNA-seq data that adjusts for the imperfect capture efficiency by estimating the number of molecules pre-dropout. Using simulated and real datasets, we saw the performance of our method is better compared to previously published methods, especially for detecting DE genes with low abundance. DECENT uses raw UMI-count data as input and does not require spike-ins, but when spike-ins are available, they can be used to improve its performance.