Increase of efficiency (η) and decrease of the required NPSH simultaneously are important objectives in the design of centrifugal pumps. In the present study, multi-objective optimization of centrifugal pumps is performed in three steps. In the first step, η and NPSHr in a set of centrifugal pumps are numerically investigated using commercial software NUMECA. Two meta-models based on the evolved Group Method of Data Handling (GMDH) type neural networks are obtained. The second step is the modeling of η and NPSHr with respect to geometrical design variables. Finally, using obtained polynomial neural networks, multi-objective genetic algorithms are used for Pareto based optimization of centrifugal pumps considering two conflicting objectives, η and NPSHr. It is shown that some interesting and important relationships as useful optimal design principles involved in the performance of centrifugal pumps can be discovered by Pareto based multi-objective optimization of the obtained polynomial meta-models representing their η and NPSHr characteristics. Such important optimal principles would not have been obtained without the use of both GMDH type neural network modeling and the Pareto optimization approach.