In the food industry, many industrial processes involve the grinding of vegetable products such as grains and seeds. The ability to be conveniently ground is an element of the quality of such products. We attempted to describe the grinding of pea seeds, taken as an example, as a discrete time Markovian process. The pea samples were passed several times through a laboratory grinder from which the screen was removed. After each series of passages, the particle site distribution of the ground products was assessed by sieving. The transition matrix of the Markovian process was estimated from these particle size distributions by using an original method based on a linear neural network.