Efficiently identifying and quantifying disease- or treatment-related changes in the abundance of proteins is an important area of research for the pharmaceutical industry. Here we describe an automated, label-free method for finding differences in complex mixtures using complete LC-MS data sets, rather than subsets of extracted peaks or features. The method selectively finds statistically significant differences in the intensity of both high-abundance and low-abundance ions, accounting for the variability of measured intensities and the fact that true differences will persist in time. The method was used to compare two complex peptide mixtures with known peptide differences. This controlled experiment allowed us to assess the validity of each difference found and so to analyze the method's sensitivity and specificity. The method detects both presence versus absence and a 2-fold change in peptide concentration near the limit of detection of the instrument used, where chromatographic peaks may not be sufficiently well defined to be detected in individual samples. The method is more sensitive and gives fewer false positives than subtractive methods that ignore signal variability. Differential mass spectrometry combined with targeted MS/MS analysis of only identified differences may save both computation time and human effort compared to shotgun proteomics approaches.