Shortened product life cycles result in increasing number of obsolete products, with adverse environmental impact. Manufacturers are facing increasing pressures from consumers and government regulators to become environmentally responsible, and have begun to setup networks to implement product recovery. This paper proposes an approach to design reverse logistics network for discrete product recovery, considering multiple objective functions (maximising net revenue and minimising environmental impact), multi-period planning horizon and uncertainty. The approach makes use of both optimisation and simulation models: mixed integer programming (MIP) to model the multi-objective, multi-period problem of network design, and simulation to handle uncertainty. Spanning-tree based genetic algorithms are utilised to find non-dominated solutions for the multi-objective model, and preferred non-dominated solutions are re-evaluated under several scenarios of uncertainty to determine the best-preferred network design. The approach is applied to a case study of part recovery implementation at a computer manufacturer.