In this talk, I discuss the construction of a multivariate generalisation of the Dirichlet-multinomial distribution (DM). The DM distribution is used in forensic genetics to compensate for subpopulation structures when evaluating the weight of evidence. This is often referred to as the theta-correction and was first introduced in the forensic literature by Balding and Nichols (1994). In order to use the theta-correction in combination with Bayesian networks (Cowell et al, 2015) we needed a generalisation of the DM distribution. Due to its properties we named this distribution the multivariate Dirichlet-multinomial distribution (MDM). By numerical examples, I show how the theta-correction affects the weight of evidence for DNA mixtures.