The shape of a trapezoidal obstacle immersed in a 2D unsteady, viscous flow is optimized by response surface (RS) techniques based on combined criteria of minimum total drag and maximum mixing efficacy. Time-dependent Navier-Stokes computations are conducted to supply the database. In order to address the issues related to noise, an outlier analysis based on iteratively re-weighted least square (IRLS) method is applied. The results indicate that optimum designs having a low mean drag coefficient tend to be square-shaped, while designs having a large value of the mixing effectiveness are more trapezoidally-shaped. Both RS and IRLS models yield consistent designs, indicating that the present task is well handled by the techniques employed. In addition, the RS methodology is used to identify domains within the design space within which all designs are, for practical purpose, acceptable.