Although current dietary guidelines focus on a combination of specific nutrients and food items, most effective dietary interventions focus on patterns of dietary intake and take into account the relationships among nutritional factors. In a controlled nutrition intervention conducted at 16 workplaces, a self-administered health habits questionnaire (HHQ) including a 67-item version of a semiquantitative food frequency questionnaire (FFQ) was distributed prior to a 15-month intervention and again after it. Principal components analysis (PCA) was used to reduce this large set of highly correlated FFQ food items to a smaller set of maximally uncorrelated components (PCs). Of the eight discrete food-based eating patterns targeted in the Treatwell intervention, six were highly correlated ([r[ > or = 0.48) with at least one PC each. This indicates a high level of concordance between a priori intervention targets and actual behavior. Based on log-transformed preintervention FFQ measures, our results showed that a very high proportion (0.55) of the variance in the FFQ data was explained by the PCs. A significantly greater increase in consumption of total vegetables and a larger decrease in dietary intake of ground and processed meats were observed among intervention companies. A comparison PCA conducted on intervention and control companies after the intervention indicated that patterns of intake were very stable over time.