In biological dosimetry by dicentric analysis, an exposure to radiation is considered non-homogeneous if the dicentric cell distribution shows overdispersion with respect to Poisson distribution. Traditionally, when this occurs, all non-homogeneous exposures are considered as partial-body exposures, assuming that there is only a mixture of irradiated and nonirradiated cells. The methods to estimate the dose in the irradiated fraction and the initial fraction of irradiated cells are based on separating which part of the cells without aberrations comes from the nonirradiated or irradiated fractions. In this study we show a new approach based on a mixed Poisson model, which allows for a distinction to be made between partial and heterogeneous exposures. To validate this approach blood samples from two donors, a male and a female, irradiated at different doses, were mixed at a 1:1 proportion to simulate partial and heterogeneous exposures. The results show a good agreement between the observed proportion of male and female cells and the proportion estimated by the model. Additionally, a good agreement was observed between the delivered doses, the initial fraction of cells and the ones estimated by the model. This good agreement was also observed after very high-dose irradiation (up to 17 Gy), when the lymphocyte cultures were treated with caffeine. Based on these results, we propose the use of this mixed Poisson model for a more accurate assessment of non-homogeneous exposures.