Nutrition education messages about eight different foods, four of them genuine and four false, were set in a Latin square and eight questionnaires were produced and distributed to customers sitting in a cafe. For each message, subjects were asked to score the extent of their agreement and compliance. Scores were adjusted to remove effects of sex, age group and social class. Correlation coefficients between agreement and compliance scores were significant (P < 0.001) for each message. Subjects who had correct knowledge tended towards correct behaviour and those who agreed with misconceptions tended to have incorrect behaviour.