It is customary to use more than one stage of flotation to obtain an acceptable level of separation of valuables. Flotation circuit design is usually accomplished by applying empirical rules or by using expertise of practitioners which may result in operations not working at their optimum conditions. Given the capabilities of genetic algorithms (GA) in finding an optimum solution in very disturbed search spaces, they can be used to obtain the desired flotation circuit configuration. In flotation circuit configuration optimization problem, a combination of metallurgical parameters such as yield and concentrate ash content could be used as the fitness function for the genetic algorithm. The multi-objective nature of the problem justified the use of the Pareto method to arrive at a set of solutions. The appropriate configuration based on technical or economic considerations could then be chosen. To obtain the fitness function for any given configuration, it is necessary to model every flotation stage. The proposed method was used to find the optimum circuit configuration for a coal washing plant. The objective was to arrive at the highest yield while producing a concentrate with a certain ash content (11.2%). The feed to the flotation circuit was characterized based on the size fractions and their flotation rate constants. Results showed that with a 95% confidence the absolute difference between the modeled and measured values were 2.9-5.5% for yield and 0.4-1.1% for concentrate ash content. When the proposed GA-based circuit configuration was implemented in the plant, the yield increased from the original value of 57.6% to 64.3% while producing concentrate ash content (10.9%) within acceptable limits. By adding one stage to the current three-stage circuit, it was predicted that the yield could be further increased by 3.8% while keeping the quality of the concentrate within the appropriate level. (C) 2013 Elsevier B.V. All rights reserved.