Updated FTP location. Recent advances in liquid chromatography/mass spectrometry (LC/MS) technology have improved the sensitivity, resolution, and speed of proteome analysis, resulting in demand for sophisticated algorithms to interpret complex mass spectrograms. Here, we propose a novel statistical method, proteomic mass spectrogram decomposition (ProtMSD), for joint identification and quantification of peptides and proteins. Given the proteomic mass spectrogram and the reference mass spectra of possible peptide ions associated with proteins from selected run or runs as a dictionary, ProtMSD estimates the chromatograms of those peptide ions, under a group sparsity constraint without using the conventional careful pre-processing (e.g., thresholding and peak picking). We show that the method was significantly improved by using protein-peptide hierarchical relationships, isotopic distribution profiles, reference retention times of peptide ions, and pre-learned mass spectra of noise. We examined the concept of database search, library search, and match-between-runs. This is the first attempt to use a matrix decomposition technique as a tool for LC/MS-based proteome identification and quantification.