According to the linguistic category model (LCM), behaviour can be described at concrete (e.g. ‘Kath hit Kim’) and abstract (e.g. ‘Kath is aggressive’) levels. Variations in these levels convey information about the person being described and the relationship between that person and the describer. In the current research, we examined the power of language abstraction to create impressions of describers themselves. Results show that describers are seen as less likeable when they use abstract (vs. concrete) language to describe the negative actions of others. Conversely, impressions of describers are more favourable when they opt for abstract descriptions of others' positive behaviours. This effect is partially mediated by the attribution of a communicative agenda to describers. By virtue of these attributional implications, language abstraction is an impression formation device that can impact on the reputation of describers. Copyright © 2009 John Wiley & Sons, Ltd.