Traffic crash hot spot analyses allow identification of roadway segments that may be of safety concern. Understanding geographic patterns of existing motor vehicle crashes is one of the primary steps for geostatistical-based hot spot analysis. Much of the current literature, however, has not paid particular attention to differentiating among cluster types based on crash severity levels. This study aims at building a framework for identifying significant spatial clustering patterns characterized by crash severity and analyzing identified clusters quantitatively. A case study using an integrated method of network-based local spatial autocorrelation and the Kernel density estimation method revealed a strong spatial relationship between crash severity clusters and geographic regions. In addition, the total aggregated distance and the density of identified clusters obtained from density estimation allowed a quantitative analysis for each cluster. The contribution of this research is incorporating crash severity into hot spot analysis thereby allowing more informed decision making with respect to highway safety.