Two experiments examined whether novel, minimal ingroups are automatically associated with positive affect while outgroups do not elicit such positive evaluative default. Participants were assigned to social categories in a typical minimal group setting and subsequently administered a masked priming task, i.e. prime words were not consciously recognized. Following either the presentation of a priori positive or negative words or the presentation of the group labels, participants classified adjectives with regard to their valence (positive/negative). In Experiment 1, a standard affective priming paradigm was realized with response latencies as dependent measures; in Experiment 2, a response window technique was used, with errors as crucial measure. In both studies, significant affective congruency effects emerged similarly for standard primes and category labels, indicating ingroup bias on an implicit level. Copyright © 1999 John Wiley & Sons, Ltd.