Labelling-based proteomics is a powerful method for detection of differentially expressed proteins (DEPs) between biological samples. The current data analysis platform relies on protein-level ratios, where peptide-level ratios are averaged to yield a single summary ratio for each protein. In shotgun proteomics, however, some proteins are quantified with more peptides than others, and this reproducibility information is incorporated into the differential expression (DE) analysis. Here we propose a novel probabilistic framework EBprot that directly models the peptide-to-protein hierarchy and rewards the proteins with reproducible quantification over multiple peptides. To evaluate its performance with known DE states, we first verified that the peptide-level analysis of EBprot provides more accurate estimation of the false discovery rates and better receiver-operating characteristic than other protein ratio analyses using simulation datasets, and confirmed the superior classification performance in a UPS1 mixture spike-in dataset. To illustrate the performance of EBprot in realistic applications, we applied EBprot to a SILAC dataset for lung cancer subtype analysis and an iTRAQ dataset for time course phosphoproteome analysis of EGF-stimulated HeLa cells, each featuring a different experimental design. Through these various examples, we show that the peptide-level analysis of EBprot provides a competitive advantage over alternative methods for the DE analysis of labelling-based quantitative datasets.