The definition of ore and waste blocks depends on selected cut off grades. This approach, even though implicitly considerers ore recovery, overlooks block performance in the plant, which can introduce errors in classification. This problem can be minimized by taking advantage of geometallurgical samples knowledge. Lab tests provide data to decide the rock category (ore, marginal ore, or waste) by looking directly at the metallurgical response. However, the category definition in the block model is not that simple and can be addressed by geostatistical methods to interpolate geometallurgical results spatially. This paper presents an example of an application in a Brazilian phosphate mine, where the ore is processed through desliming, demagnetization, and flotation to obtain a phosphate concentrate. The definition of ore/waste blocks is actually made considering only P2O5 and CaO grades, which works well for the isalterite horizon but brings problems when applied to the other weathering profiles. The solution used to incorporate geometallurgical information in this decision-making process applies hierarchical indicator kriging to assign a category to each block: ore (category 1), marginal ore (category 2), or waste (category 3). Next, the blocks classified as categories 1 or 2 received estimations of yield and recovery of P2O5 in the flotation concentrate and tailings using neural networks. Finally, it was possible to evaluate that if the old criterion was applied, more than 29% of the blocks in the semi-weathered horizon would be misclassified as waste, even presenting a high yield in flotation concentrate. This study shows the importance of incorporating geometallurgical information in an ore block definition, reducing errors, and improving mining profit.