One of the most intriguing aspects of Fishbein's (e.g. 1980) theory of reasoned action is the logic of the double negative: if a behaviour is thought to be unlikely to result in a negatively evaluated consequence, then the product of the two negatives (an unlikely consequence that is negatively valued) is thought to provide an impetus for the formation of a positive attitude towards the behaviour. The present experiment tested two derivations from the logic of the double negative. First, according to this logic, whether beliefs and evaluations are positively or negatively framed should not affect their ability to predict attitudes. Second, the multiplicative assumption of this logic (a negative × negative = a positive) suggests that a multiplicative model should be a superior predictor of attitudes compared to an additive model that does not allow for the logic of the double negative. The obtained findings contradicted both of these predictions.