ABSTRACT.- For models with several base populations (Unknown Parent Groups or Metafounders), the usual definition of reliability is ill-posed. Here we propose to define reliability based on contrasts with one or several metafounders, leading to a sounder genetic interpretation. In the case of a single metafounder, our definition equals the definition of reliability for the classical animal model. This definition also allows expressing the reliability of contrasts of metafounders, which may be of interest to decide if their setup is estimable with sufficient reliability. All desired quantities can be obtained from elements of the inverse of the MME.