The spatial variability in crop yields commonly observed in smallholder farms of sub-Saharan Africa is often caused by gradients of declining soil fertility with increasing distance from the homestead. This heterogeneity means that recommendations based on regional soil surveys are of limited value. The variability in soil qualities within farms must be considered when designing management strategies, and their feasibility analysed by integrating results at the farm livelihood scale. For this purpose, we have developed the model FARMSIM, a dynamic bio-economic model for analysis and exploration of trade-offs in resource and labour allocation in heterogeneous smallholder farms. Focusing on farm-scale strategies, the approach to simulation of soil and crop processes in FARMSIM (the sub-model FIELD) is designed to be simple, but to keep the necessary degree of complexity to capture heterogeneity in resource use efficiencies. To test our approach, the sub-model FIELD was calibrated against chronosequences of woodland clearance in three agroecological zones of Zimbabwe (with soil textures of 3, 10, 35% clay), and used to simulate: (i) the creation of soil fertility gradients, and (ii) different strategies of N, P and manure applications to maize and soyabean rotations in homefields and outfields of smallholder farms on clayey and sandy soils. The results of the simulation of management strategies were tested against on-farm experimental data from Murewa, Zimbabwe. The model produced satisfactory predictions (r2: 0.6-0.9) of long-term changes in soil organic C, of crop responses to N and P and of nutrient use efficiencies across a wide range of yields and different field types. This demonstrated the broad applicability of the model despite the sparse data required for initialisation. However, the model results were less accurate in predicting crop responses to N and P applications in the outfields on sandy soils. Experimental evidence indicated yield limitation by Ca and