Multiobjective evolutionary algorithms have already been used for the optimization of the processing conditions of a co-rotating twin screw extruder. In the present paper, we develop this approach for the optimization of the screw geometry, in terms of the most adequate location of a pre-selected set of screw elements, in order to maximize a prescribed performance. Experimental validation shows the pertinence of this approach, which is computationally more rapid and less expensive than the classical trial and error procedure. An application to the complex case of reactive extrusion is also presented.