Three samples, consisting of 200 traits, 200 nouns and 200 verbs taken from the Toglia et al. (1978) verbal norms, were analysed for evidence of positive-negative asymmetry. Within each sample the items were ordered on a general positive-negative index, and a systematic series of dichotomizations was carried out, ranging from S per cent negative to 95 per cent negative. It was found that the partial correlation of pleasingness with the positive-negative dichotomization increased as a linear function of the informational complexity of the negative class of words, for all three word types. The partial correlation of familiarity increased as a similar function of the positive class of words, but only for traits. The results are discussed in terms of Peeters and Czapinski's (1990) positive-negative asymmetry model, and Scherer's (1984) component model of emotion.