The behaviour of animals towards their mirror image ("mirror test") is routinely used as a proxy to measure aggression levels, especially in fish. The lack of evidence for visual self-recognition in fish supports this method. However, recent work points towards different hormonal and gene expression responses when fish are exposed either to conspecific opponents or to their mirror image, urging for validation of this widespread method. Here, we test the predictive value of mirror tests in three sympatric cichlid species from Lake Tanganyika: the cooperative breeder Neolamprologus pulcher, the polygamous shell brooder Telmatochromis vittatus and the monogamous, biparental piscivore Lepidiolamprologus elongatus. In particular, we compare differences in restrained and overt aggression levels for individuals of each species when confronted with a mirror or a live conspecific. The three species differed in response to the two contest situations. While in N. pulcher both aggressive responses were correlated between the mirror test and the live opponent fight, there was no such relationship in T. vittatus and L. elongatus. Thus, the mirror test appears to be a suitable surrogate for intraspecific aggression in N. pulcher, while aggression against a mirror image has limited predictive value for intraspecific aggression in the other two species. These results underline the importance of validating the mirror test's predictive value in a study species before drawing conclusions from mirror tests about aggressiveness under natural, social conditions.