This study applied a data mining process to the French Guide for Road Earthworks (GTR), aiming to predict the capacity of compaction and the compaction condition parameters. The database is comprised mainly by the data available in the "Compaction Tables", part of GTR. The data includes qualitative variables (material, compactor type and energy level) and quantitative variables (Q/S parameter, layer thickness and velocity, number of load applications and theoretical compaction capacity of the compactor). Computations were carried out using the R tool and RMiner library. The best models obtained, based on neural networks and support vector machines, allowed the finding of qualitative and quantitative relationships between variables involved in compaction. Furthermore, a prototype was developed that allows an automatization of the information of the compaction conditions. © 2009 Taylor & Francis Group, London.