Correspondence studies of labour market discrimination find that minorities, which in general suffer disadvantage, are sometimes preferred in a choice against members of the majority. This outcome has been observed in several studies of ethnic or nationality-based discrimination, but also in studies focusing on other characteristics, such as unemployment and being overweight. However, it is generally not explained and dismissed as noise. In this paper we challenge this understanding, and, using meta-analytical techniques, we show that instances of minority preference are not randomly distributed. We also show that they are more frequent for groups which overall suffer stronger discrimination and for high skilled professionals. We reason that this result may be explained with the fact that groups that suffer discrimination have fewer alternatives in the labour market and this makes them more attractive for jobs of sub-standard quality and for jobs in which turnover costs are high (e.g. high skilled professionals). We conclude by arguing that since tests in which the minority candidate is preferred are not randomly distributed, future research should study the determinants of minority preference in a more systematic manner.