For MALDI-TOF mass spectrometry, we show that the intensity of a peptide–ion peak is directly correlated with its sequence, with the residues M, H, P, R, and L having the most substantial effect on ionization. We developed a machine learning approach that exploits this relationship to significantly improve peptide mass fingerprint (PMF) accuracy based on training data sets from both true-positive and false-positive PMF searches. The model's cross-validated accuracy in distinguishing real versus false-positive database search results is 91%, rivaling the accuracy of MS/MS-based protein identification.