Background: We aimed to illustrate the importance of imputation models specifications, based on a study exploring the associations between subclasses of dietary polyphenols and the thiobarbituric-acid-reactive substances (TBARS). Methods: Data were collected in a long-term cohort study based on a double-blind randomized placebocontrolled nutritional trial (SU.VI.MAX 2 study). The association between polyphenols intakes and TBARS were studied using linear regression models. Missing data were handled using multiple imputation with chained equations. Results: A total of 4,129 subjects were included in the analysis, 2,116 of whom had an available outcome measure (TBARS). Differences in selected predictors of TBARS according to the handling of missing data on both covariates and outcome (complete case analysis or multiple imputation) were observed. In the complete case analysis, none of the dietary polyphenol subclasses was found to be associated with TBARS while based on multiple imputed datasets, two polyphenol subclasses, namely catechins and hydroxybenzoic acids, could be selected as associated with TBARS. Of note, while there was a positive association between catechins and TBARS, the hydroxybenzoic acids were negatively associated with TBARS. Conclusions: Adequate modelling of missing data on both covariates and outcome allowed dietary catechins intake to be selected as associated with a biomarker of oxidative stress.