Response surface methodology (RSM) was utilized to investigate the dose-response relationships of a phenolic mixture (catechin, genistein and daidzein) as a pre-thermal processing technique to reduce reactive carbonyl species (RCSs; glyoxal, methylglyoxal and 3-deoxyglucosone) in ultra-high temperature (UHT) bovine milk. A modified derivatization technique for RCSs was developed to overcome quantitative error caused by interference from the phenolic compounds. For the statistical analysis, a Box-Behnken 3-factor (catechin, genistein and daidzein) 3-level (0.17, 0.645 and 1.12 mM) design was employed. In general, all phenolic mixtures were able to reduce RCSs in UHT milk; some compositions reported RCSs levels at or below levels reported in pasteurized milk. Predictive models with no significant lack of fit (p > 0.05), high R(2)-values (0.886-0.979) and good predictive power were developed. ANOVA analysis of the glyoxal levels indicated that only linear effects of each phenolic compound had a significant effect (p < 0.05) meaning that no significant interactions between the different phenolic compounds influenced glyoxal levels. Linear, cross product and quadratic effects of factors were reported (p < 0.05) for methylglyoxal, indicating more complicated interactions between the phenolic compounds. Both linear and quadratic effects were also reported (p < 0.05) for 3-deoxyglucosone. Overall, based on canonical analysis, catechin seemed to be the most influential factor for the reduction of RCSs in UHT milk. In summary, RSM provided a basis to understand phenolic structure-reactivity and to optimize the composition of a tertiary mixture of phenolic compounds for reduction of RCSs in UHT milk.