Since crash occurrence are typically aggregated as clusters in space and crash data are always collected with certain spatial scale, intrinsic spatial effects exist extensively in road safety analysis. It is a common issue for crash prediction models and has gained lots of focus in the past decades toward different aspects. Therefore, having a comprehensive understanding of spatially distributed crash data is indispensable for safety inspection, and incorporating spatial effects in both micro-level and macro-level crash prediction modeling is expected to represent the true underlying data generating processes. This paper provides a detailed review of the spatial data characteristics in road safety analysis, including three key issues, i.e., multilevel data structure, spatial dependence and heterogeneity, as well as methodological approaches that have been used to solve these problems. Whereas these spatial analysis technique substantially improved the accuracy and robustness of crash prediction, zonal level practice has been highlighted in this study to specify the use of these methods.