Tolerance design provides a common ground for design and manufacturing engineers. It has significant impacts on the quality, cost, and manufacturing lead time of a product. Most of the literature treats tolerances as continuous, thus continuous analytic models of various types are established based upon the discrete tolerances and manufacturing data with the regression method. It’ll be demonstrated in this research that this method will lead to solutions of both mathematical suboptimum and manufacturing infeasibility. This paper proposes using discrete tolerances with an integer programming approach so that solutions of both mathematical optimum and manufacturing feasibility are ensured at the design stage of the product. Numerical examples illustrate the differences between the discrete and continuous tolerance approaches.