Computational models based on 3D meshes are ubiquitous in are such as game, animations and virtual reality. However, very large data sets are frequently produced, e.g. by scanners 3D and fluid dynamics simulations, wich require high computer power to be handled. Mesh simplification tecniques, preserving the topology and the geometry of the mesh, are then implemented to bring the datea to a size suited to be used in such areas. In this work we introduce a new tecnique wich we call Markov Geometric Diffusion based on probability transition matrix tecniques and built upon a data set organized geometricallyas a mesh. This method puts together a strategy based on a geometrically constructed Markov chain, wich control, in a probabilistic way, a normal vector field to the mesh, with a simplification method capable of estimating the impact of element removal in the mesh structure. Several error evaluation metrics are used tocompare the error of the simplified mesh with the original one.