The current heat transfer and fluid flow models of friction stir welding can predict temperatures, material flow, and residual stresses from welding parameters such as the welding velocity, tool rotational speed and axial pressure. Since these models are unidirectional, they cannot currently predict welding variables needed to attain a thermal cycle or other weld attributes. Here we show that a differential evolution based optimization technique can be used to achieve inverse modeling capability of friction stir welding models. The new capability enables systematic tailoring of weld attributes based on scientific principles without any time consuming trial and error approach. The inverse modeling capability is tested by comparing model predictions with experimental results.