Abstarct: Many experimental studies need to use variables that are influenced by their geographical locations. In such cases, to consider the spatial structure of data are essential and ignoring this structure results in to lose some relevant information. To deal with such data, the classical linear regression model is not efficient. Indeed, the spatial structure leads to the violation of homogeneity and uncorrelation assumptions. Hence, we should use some appropriate spatial models. To have sufficient flexibility, we need a model that considers both spatial heterogeneity and dependency, simultaneously. However, the real heterogeneity and dependency structure of data is usually unknown. ‎Furthermore, there is no available prior information about this structure. ‎‎ In this thesis, we use an iterative locally weighted regression approach to determine homogenous spatial regions, named as spatial regimes. Next, we apply these regimes in a class of spatial autoregressive models, known as spatial econometric models. We compare the proposed models with the linear regression model and classical spatial econometric models baxsed on the Akaike information criterion. The results show a better performance of econometric models with the endogenous spatial regimes.