Flavour preference learning in 21 restrained and 21 unrestrained females was explored using an evaluative conditioning paradigm. Each participant was exposed to an adapted version of the procedure used by Johnsrude et al., (1999, Learning & Motivation 30, 250-264). During conditioning, participants sampled 10 instances each of three novel flavours presented in a semi-randomized order. After sampling a flavour they were instructed to eat a sweet according to three different flavour-reinforcement contingencies. One flavour was accompanied by a sweet on 90% of trials and was presented alone on 10% of the trials, while the remaining flavours were rewarded at ratios 50% :50% and 10%: 90%, respectively. The conditioning phase occurred in conjunction with a counting task requiring continuous use of working memory. and was immediately followed by the participants making hedonic ratings of each flavour. Very few participants showed awareness of the purpose of the experiment or the specific reward contingencies. Despite this, the ratings given by the unrestrained caters were highly correlated with the reward ratio experienced during conditioning. In contrast, restrained caters exhibited no evidence for evaluative learning. These findings may explain the equivocal nature of results from previous studies of positive flavour-flavour learning (FFL) and may offer a novel theoretical context within which to study dietary restraint. (C) 2001 Academic Press.