In this paper, an optimal design of a three-prismatic-universal-universal (3-PUU) translational parallel manipulator is performed using the genetic algorithm (GA) to achieve optimum stiffness characteristics. The stiffness matrix is derived intuitively based upon an overall Jacobian with the consideration of actuation and constraints, and the compliance subjected to both actuators and legs is taken into account to establish the stiffness model. A global stiffness index defined as the mean value of the determinant of stiffness matrix over the reachable workspace is adopted as a stiffness measure, meanwhile the architectural parameters are optimized via a GA approach. The research results are valuable in the design of a 3-PUU parallel manipulator for machine tool applications.