Soils play a crucial role in nutrient cycles where they act as source and sink. In the tropics, the genesis and functioning of soils are not fully understood yet. Many variables interact and contribute to the spatial variability of soil properties. The objective of our study was to test if soil-landscape modeling is an appropriate technique to predict the spatial variability of total and exchangeable K and Mg in a humid tropical forest in Panama in order to contribute to our understanding of soil-landscape processes in the tropics. We used a design-based, stratified sampling scheme to collect soil samples at 108 sites on Barro Colorado Island, Panama, which has an area of 1500 ha. Stratifying variables are lithology, vegetation and topography. Topographic variables were generated from a high resolution digital elevation model with a grid size of 5 m.  We took samples from 5 depths down to one meter, and analyzed for total and exchangeable potassium and magnesium. Classification and Regression Trees (CART) were then adopted to develop models that regionalize the point observations using digital terrain data as explanatory variables. Our results suggest that the selected predictors do not control the spatial distribution of the exchangeable K and Mg in the soil. Therefore, neither geomorphic nor hydrologic processes account for the spatial variation of soil exchangeable K and Mg.