ABSTRACT.- Dry forests are among the most threatened ecosystems globally, due to agricultural expansion driven by the increasing demand for food, fibers, and energy in developed and emerging countries. Among these, the forests of the South American Gran Chaco are one of the global deforestation hotspots. The Argentine Dry Chaco has been the focus of several studies that assess the factors that drive forest conversion. However, these studies do not describe the causal relationships among these drivers and seldom use existing theory to select drivers. Here we employ a theory-driven approach to test the relative merits of alternative and complementary hypotheses to explain the drivers and mechanisms explaining the unequal spatial distribution of forest loss and maintenance in the Argentine Dry Chaco from 2000 to 2010. Using structural equation modeling, we quantified the direct and indirect effects of multiple drivers and compared the explanatory power and parsimony of these alternative hypotheses, i.e. the biophysical, infrastructure, socio-demographic, institutional, and the integration of them. For both forest loss and maintenance, the model containing infrastructural drivers had the best balance between parsimony and explanatory power. Integrated models, comprising a combination of drivers, had the highest explanatory power (R2 = 0.81 for forest maintenance, and R2 = 0.58 for forest loss). We show that biophysical constraints operate directly and indirectly:soil suitability had direct effects on forest cover maintenance, while precipitation affected it both directly and indirectly through influencing the institutional (land tenure) and infrastructure (road density). Indigenous communities positively affected forest maintenance both directly and indirectly mediated by non-private land tenure. Our results suggest that disentangling the structure of the relationships among drivers could increase our capacity for understanding and steering land-use change. Furthermore, policies for halting deforestation might increase their effectiveness by accounting for the mechanisms that underlie forest loss and maintenance. © 2021 Elsevier Ltd