From social identity theory a negative relation between self-esteem and ingroup bias can be deducted. Much research has been done to test this proposition and largely failed to confirm this relation. Unlike many existing studies, we conducted an experiment in which (a) self-esteem is not conceived as a trait entity but much more situation-specific, (b) the self-esteem manipulation is not relative to the outgroup, and (c) the measure of intergroup differentiation is unrelated to the self-esteem manipulation. We categorised our participants into two arbitrary minimal groups (Klee or Kandinsky fans) and afterwards formed homogeneous three-person groups (all persons were either Klee or Kandinsky fans). We manipulated the state self-esteem of these real groups by giving them positive or negative feedback concerning their performance in a problem-solving task. Afterwards, all groups distributed money to ingroup and outgroup members via Tajfel distribution matrices. Low state self-esteem groups were found to exhibit stronger ingroup bias than high state self-esteem groups overall, although the variability of intergroup discrimination was larger in the low state self-esteem groups, pointing to more heterogeneous reactions to low state self-esteem.