In the context of rapidly growing African cities, a thorough understanding of the complexities of urban food systems is essential for addressing the challenges of food insecurity and undernourishment for city dwellers. Particularly in South Africa, where pre-existing inequalities drive disparities in food access and diet-related health outcomes, a comprehensive perspective including the spatial distribution of malnutrition in urban environments is required to develop effective interventions. The present study examines the essential elements of an urban food system by employing a Bayesian network as a causal framework. By integrating survey data from households and food outlets with spatial information, a food systems model was created to test policy interventions. The study demonstrates the challenges of intervening in complex urban food systems, where dietary choices are shaped by various factors, often in a spatially heterogeneous manner. Interventions do not always benefit the targeted groups and are sometimes ineffective as result of system interactions. Our study shows that Bayesian network models provide a powerful tool to effectively analyse the complex interactions within such systems, thereby enabling the identification of optimal combinations of multifactor interventions. In our case study for Worcester, South Africa, the results reveal that the largest potential for improvement of food and nutrition security lies in the informal food sector, and support for affordable and local fresh produce is a viable measure for enhancing local nutrition, though the extent of impact varies across the city.