Many fish species are social, and individuals spend most of their lives in shoals, but sociability can vary greatly between species, populations and individuals. Sociability has been largely studied by measuring the time spent by a focal fish in proximity to one or more conspecifics. To control for the behaviour of the stimulus fish, the conspecifics have often been substituted by a mirror, on the assumption that the subject perceives its mirror image as a conspecific. The validity of the mirror test has recently been questioned, at both the behavioural and the molecular level, because of the discrepancy in fish responses when exposed to a mirror image and to a live conspecific. In this study, we compared the sociability scores of a social fish, the guppy, Poecilia reticulata, obtained using live fish or a mirror as a stimulus, to assess the validity of the mirror test. We found that the sociability score assessed using the standard mirror test was not significantly correlated with the sociability assessed using live stimuli. Nevertheless, we observed a positive correlation between the scores of the two tests when the mirror test was performed in a more naturalistic context, controlling for the minimum distance between the stimulus fish and the mirror. Our findings provide evidence of the reliability of the mirror test as a measure of sociability compared to the test using live conspecifics when certain requirements are satisfied.