This paper presents a method for the capacity planning of a manufacturing system subject to significant uncertainty, both in the demand and in the process. The uncertainty is modeled using phase-type distributions. By fitting this type of distributions to the empirical data, we can model the system as a Markov chain in other to estimate the system performance measures for each given system configuration. Evaluating different system configurations we obtain the estimated trade-off functions relating performance with capacity. Using these functions we select using a mathematical program the appropriated capacity, taking into account the system objectives. The method is applied to a real case following a make-to-order job flow policy.  The problem is this case is how to decide on the utilization of the two different resources available for a given lead-time. The solution obtained with the method proposed is compared with the deterministic approach solution.