One of the most important objectives in a fed-batch operation is that of finding the input feed profile that optimizes the production/consumption of a desired component. If systems are better represented in a stochastic framework, as may be the case with bioprocesses, an approach different from classical optimization based on differential equations, can be used. In this work, a stochastic input/output model is used to describe the fed-batch fermentation for ethanol production. A transformation of the model into a Markov chain is made for posterior use in the problem definition of the Markov Decision Process aimed at maximizing the ethanol production. The results obtained from a real time experiment show satisfactory results.