As in most European countries, traffic safety has become top priority in the National Safety Plan in Belgium. The first phase in every safety analysis concerns the identification of the hazardous locations. In this respect, a local indicator of spatial association (Moran’s I) is improved and applied to determine hot spots locations on highways in Limburg, a province in Belgium. However, the analysis is complicated by the fact that accident data have a very specific nature: they form a Poisson random process rather than a Gaussian random process and they are prone to sparseness. Therefore, the well-established indicator needs some adaptations and simulations are required to determine the underlying distribution of Moran’s I. This paper emphasizes the importance of using a correct distribution and indicates what could go wrong otherwise e.g. at policy level.