A Monte-Carlo code has been developed to infer dissolution rates of silicate glasses from their chemical composition. The glasses are considered as solid solutions of oxides and silicates, being divided into elementary cells of network formers, in which network modifiers can be incorporated. The different types of cells are randomly distributed and a resistance capacity to dissolution is assigned to each cell. We have tested the case where each cell is removed in a constant time as soon as it is in contact with the solution and a second one where each cell is dissolved in a time depending on the number of already dissolved neighbor cells. The first hypothesis leads to a percolation threshold when the proportion of fast etching cells reaches 25 %. Above, the dissolution front progresses via contiguous channels of fast etching cells. In the second hypothesis the variations of the etch rate with the proportions of each component are much closer to a linear combination of individual etch rates. These predictions have been successfully compared to experimental data on different series of silicate glasses of increasing complexity. Such a treatment could be straightforwardly extended to nuclear glasses.