Different spatial configurations (or scenarios) of multiple best management practices (BMPs) at the watershed scale may have significantly different environmental effectiveness, economic efficiency, and practicality for integrated watershed management. Several types of spatial configuration units, which have resulted from the spatial discretization of a watershed at different levels and used to allocate BMPs spatially to form an individual BMP scenario, have been proposed for BMP scenarios optimization, such as the hydrologic response unit (HRU) etc. However, a comparison among the main types of spatial configuration units for BMP scenarios optimization based on the same one watershed model for an area is still lacking. This paper investigated and compared the effects of four main types of spatial configuration units for BMP scenarios optimization, i.e., HRUs, spatially explicit HRUs, hydrologically connected fields, and slope position units (i.e., landform positions at hillslope scale). The BMP scenarios optimization was conducted based on a fully distributed watershed modeling framework named the Spatially Explicit Integrated Modeling System (SEIMS) and an intelligent optimization algorithm (i.e., NSGA-II, short for Non-dominated Sorting Genetic Algorithm II). Different kinds of expert knowledge were considered during the BMP scenarios optimization, including without any knowledge used, using knowledge on suitable landuse types/slope positions of individual BMPs, knowledge of upstream–downstream relationships, and knowledge on the spatial relationships between BMPs and spatial positions along the hillslope. The results showed that the more expert knowledge considered, the better the comprehensive cost-effectiveness and practicality of the optimized BMP scenarios, and the better the optimizing efficiency. Thus, the spatial configuration units that support the representation of expert knowledge on the spatial relationships between BMPs and spatial positions (i.e., hydrologically connected fields and slope position units) are considered to be the most effective spatial configuration units for BMP scenarios optimization, especially when slope position units are adopted together with knowledge on the spatial relationships between BMPs and slope positions along a hillslope.