Under favorable conditions, diseases can cause losses up to 100% in cacao. Breeding has been the most effective strategy in reducing these losses. However, breeding is limited by the: a) long generation time; b) large tree size (9 m2/tree); c) multiple measurements per year; d) low heritability; e) requirement of the presence of the pathogen; among others. Genomic selection (GS) overcomes some of these problems. In 2011, a project on genomic selection involving Ceplac and Cirad was started, aiming to implement GS on breeding for diseases resistance and yield. In 2013, a second project is being started involving Ceplac, Cirad and Catie, aiming preventive breeding for moniliasis (a disease still absent in Brazil) resistance through GS. In the first project, a training population of 380 progenies, out of 4000 trees of the third cycle of Cepec´s breeding program, and 90 founder and pedigree clones were chosen. In the second project, a training population of 470 trees was chosen among 2500 trees established in progeny trials and evaluated for moniliasis at CATIE. Founder parents (resistant to moniliasis) of this population was also introduced in Brazil and involved in crosses for preventive breeding. Prediction equations developed in Costa Rica will be used in Brazil for selection in the absence of the pathogen. Aiming to have accurate "phenotype" measurements for the prediction equations, BLUP values from historical data have been used (around 800 thousand data points, of 15 thousand trees). The overall strategy and partial results of these projects will be presented. Work supported by Agropolis Foundation, Fapesb, Capes, Ceplac. (Texte intégral)