A novel flavour may become liked if it is presented repeatedly and in combination with a second flavour that is already liked. Conceptually, this 'flavour-flavour learning' is important, because it can account for many of our everyday food and flavour preferences. However, relatively little is known about the underlying process because learning paradigms have lacked reliability. Based on previous research we explored whether learning is determined by three variables; i) hunger state, ii) demand and contingency awareness, and iii) dietary restraint. Participants (male n=15/female n=15) consumed three different and novel-tasting fruit teas. One of the teas had a non-caloric sweetener added (CS+) and two were unsweetened (CS-). Before and after this training the participants ranked their preference for unsweetened versions of the three teas. We found that the training increased preference for the CS+ relative to the CS- teas. However, this effect was only found in hungry participants. We also found little evidence that learning was related to whether the participants could identify (recognition test) the specific tea that had been sweetened during training, suggesting that the underlying process is automatic and it operates outside conscious awareness. Learning was not predicted by dietary restraint (measured using the DEBQ-R scale). Together, these findings provide further evidence for a linkage between flavour-flavour learning and flavour-nutrient learning. (C) 2007 Elsevier Inc. All rights reserved.