The heterogeneous three-dimensional spatial distribution of mycotoxins has proven to be one of the main limitations for the design of effective sampling protocols. Current sample collection protocols for mycotoxins have been designed to estimate the mean concentration and fail to characterise the spatial distribution of the mycotoxin concentration due to the aggregation of the incremental samples. Geostatistical techniques have been successfully applied to overcome similar problems in many research areas. However, little work has been developed on the use of geostatistics for the design of sampling protocols for mycotoxins. This paper focuses on the analysis of the two and three-dimensional spatial structure of fumonisins B₁ (FB₁) and B₂ (FB₂) in maize in a bulk store using a geostatistical approach and on how results help determine the number and location of incremental samples to be collected. The spatial correlation between FB₁ and FB₂, as well as between the number of kernels infected and the level of contamination was investigated. For this purpose, a bed of maize was sampled at different depths to generate a unique three-dimensional data set of FB₁ and FB₂. The analysis found no clear evidence of spatial structure in either the two-dimensional or three-dimensional analyses. The number of Fusarium infected kernels was not a good indicator for the prediction of fumonisin concentration and there was no spatial correlation between the concentrations of the two fumonisins.