Abstract. We introduce a multi-objective formulation of service-oriented testing, focusing on the balance between service price and reliability. We experimented with NSGA2 for this problem, investigating the effect on performance and quality of composition size, topology and the number of services discovered. For topologies small enough for exhaustive search we found that NSGA2 finds a pareto front very near (the fronts are a Euclidean distance of ∼ 0.00024 price-reliability points apart) the true pareto front. Regarding performance, we find that composition size has the strongest effect, with smaller topologies consuming more machine time; a curious effect we believe is due to the influence of crowding distance. Regarding result quality, our results reveal that size and topology have more effect on the front found than the number of service choices discovered. As expected the cost-reliability relationship (logarithmic, linear, exponential) is replicated in the front discovered when correlation is high, but as the price-reliability correlation decreases, we find fewer solutions on the front and the front becomes less smooth.