AbstractBitterness represents a major challenge in industrial application of food protein hydrolysates or bioactive peptides and is a major factor that controls the flavor of formulated therapeutic products. The aim of this work was to apply quantitative structure‐activity relationship modeling as a tool to determine the type and position of amino acids that contribute to bitterness of di‐ and tri‐peptides. Datasets of bitter di‐ and tri‐peptides were constructed using values from available literature, followed by modeling using partial least square (PLS) regression based on the threez‐scores of 20 coded amino acids. Prediction models were validated using cross‐validation and permutation tests. Results showed that a single‐component model could explain 52 and 50% of the Y variance (bitterness threshold) of bitter di‐ and tri‐peptides, respectively. Using PLS regression coefficients, it was determined that hydrophobic amino acids at the carboxyl‐terminus and bulky amino acid residues adjacent to the carboxyl terminal are the major determinants of the intensity of bitterness of di‐ and tri‐peptides. However, there was no significant (p>0.05) correlation between bitterness of di‐ and tri‐peptides and their angiotensin I‐converting enzyme‐inhibitory properties. Copyright © 2006 European Peptid