Migration flows are affected by both origin and destination attributes as well as spatial interaction between them. However, most studies build their insights on one-step gravity spatial interaction models, which fail to distinguish the spatial interaction effect from the attributes as the dispersal capacity and the attraction capacity. In particular, the effect of spatial interaction, often approximated solely by distance and viewed as the most intractable component in the models, leads to model biases and hinders the theoretical interpretation of migration mechanisms and spatial patterns. Drawing on the method of migration flow decomposition and structural gravity model theory, this paper replaces the one-step “flow–variable” approach with a two-step “flow–effect–variable” framework for modelling migration flows. Based on interprovincial migration data from China’s Sixth and Seventh National Population Censuses, the paper decomposes migration flows to measure the effects of dispersal capacity, attraction capacity, and spatial interaction, identifying temporal trends in their effects. Building on measured effects of spatial interaction, the paper uses multiple regression analyses to identify key influencing factors of spatial interaction, and examines how bilateral explanatory variables shape the spatial structures.