A farm survey was conducted within a 100 km2 sampling block to collect data on the spatial variation in unfertilized maize biovolume and grain yields in relation to soil organic carbon (SOC), total nitrogen, phosphorus and extractable cations. Key soil factors associated with crop performance were identified using stepwise multiple linear regression modelling. The spatial variation of key soil factors and crop performance indicators (CPIs) was described in terms of spatial dependency. An analysis of variance indicated the variation explained by soil types, sampling units, and administrative units. Soil properties displayed high variability with coefficients of variation of in the range of 50% to 89% for extractable nutrients. Grain yield ranged widely from 0.1 to 11.3 t ha−1, with 31% of the variation being accounted for by measured soil properties. SOC was identified as key soil factor associated with variation in crop performance. SOC displayed moderate spatial dependency with a range of 523 m. Analysis of variation indicate that variation in SOC was sufficiently described by small spatial units (fields). These insights were used to provide a framework for determining an appropriate scale for developing digital soil maps or distance for soil sampling in heterogenous smallholder farming systems. Strategies aimed at refining fertilizer use recommendation can therefore use this guideline.