Gene tree/species tree reconciliation is a recent decisive progress in phylo-genetic methods, accounting for the possible differences between gene histories and species histories. Reconciliation consists in explaining these differences by gene-scale events such as duplication, loss, transfer, which translates mathematically into a mapping between gene tree nodes and species tree nodes or branches. Gene conversion is a very frequent biological event, which results in the replacement of a gene by a copy of another from the same species and in the same gene tree. Including this event in reconciliations has never been attempted because this changes as well the solutions as the methods to construct reconciliations. Standard algorithms based on dynamic programming become ineffective. We propose here a novel mathematical framework including gene conversion as an evolutionary event in gene tree/species tree reconciliation. We describe a randomized algorithm giving in polynomial running time a reconciliation minimizing the number of duplications, losses and conversions. We show that the space of reconciliations includes an analog of the Last Common Ancestor reconciliation, but is not limited to it. Our algorithm outputs any optimal reconciliation with non null probability. We argue that this study opens a wide research avenue on including gene conversion in reconciliation, which can be important for biology.