The Japanese authorities have been measuring since March 16 th the radiation level around the Fukushima nuclear plant as well as in the 46 other prefectures of the country. Considering these daily measurements of radiation level in the air, in fallout, tap water, dust, soil, leafy vegetables and seawater, we propose to assess the population exposure to the radioactivity by using Bayesian networks. Bayesian networks are handy thanks to their dynamic component, calculating probability distributions in accordance with the " system state " variable (in this case, the exposure of the population) over time. 1. INTRODUCTION The aim of this study is to quantify the health impact of the radioactive discharges. We will distinguish external exposure through inhalation of radionuclides from internal exposure through ingestion of contaminated water or food because internal exposure is 20 times more dangerous than an external one. Because of the uncertainty due to the health impact assessment, the methodology chosen is a probabilistic modelling with Bayesian networks. More precisely, we are going to use a Dynamic Bayesian Network (DBN) enabling us, informed of the population exposure at some point, to simulate its evolution over time and then to detect when the threshold of 1 Sievert (at which the effects on health appear) is reached. We also hope to be able to answer the following question: depending on the areas, which contamination mode could be most important (external, internal or both)? Our goal is to define the areas to which the Japanese health authorities should really pay attention in the future in order to prevent as much as possible diseases related to exposure to radionuclides such as iodine-131 or cesium-137.