When performing quantitative epidemiological analyses, particularly those linking human exposure to health stressors, population figures are one of the inputs we need at specific spatial support. However, we often need to operate with aggregated data at disparate and coarse spatial support. This situation raises the problem of producing population estimates for the desired spatial support. Additionally, we must also account for the uncertainty associated with such population estimates, along with other sources of uncertainty, in order to propagate it accordingly. The goal of this thesis is to present a statistical approach for disaggregating population counts of a defined geographical region by combining misaligned spatial data. The probabilistic framework consists of a Bayesian hierarchical model for population counts where the underlying spatial process is approximated by a Markov Random Field. Numerical methods and Laplace approximations are advocated for deriving the marginal posterior distributions of the latent population counts, which are summarised by means of estimates of expected values and variances. To elucidate the proposed statistical framework, a sequence of areal downscalers are fitted to population counts corresponding to census data of Belgium's municipalities in 2011 by using a land-cover map as ancillary data.