We propose a new method based on probe-level data (PLIDEG) to filter differentially expressed genes from non-differentially expressed genes. We compare this new method with others based on expression values by using two spikein data sets. With the extra information provided by probe level data, PLIDEG not only controls type I error, but also increases the power of detecting DEGs, simultaneously. Therefore, PLIDEG can efficiently separate DEGs and non-DEGs without requiring the estimation of the number of non-DEGs. Based on theoretical analysis and results from application to real microarray data, we confirm these good features of this new method.