There is an increasing interest in the identification and quantification of proteomic markers for biological and medical research. Experimental techniques utilizing liquid chromatography with mass spectrometry (LC- MS) provide one method of directly observing protein expression, but a substantial amount of computational work is needed to go from the raw LC-MS data to a matrix format analogous to gene expression studies in which the columns are samples and rows are proteins. One critical step in this pipeline is the extraction of pep- tide features from the LC-MS signal data. We present a complete solution to LC-MS feature detection that combines a model-based approach to feature extrac- tion on the MS scans with techniques for robust estima- tion to build LC-MS features from the individual scans. We show that using our approach, we find significantly more features, more matches, and better correlation be- tween replicated LC-MS experiments than are found us- ing the current state-of-the-art software.