Design efficiency and robustness at early stage of parametric engineering design play a critical role in reducing cycle time and improving product quality in the overall product development process. Usually, the “forward mapping” approach, is used to find designs, where the desirable performances are satisfied through large iterations of analysis and evaluation from design space to performance space. However, these approaches are time-consuming and involve blind search if the engineering system simulation models and/or initial conditions are not appropriately selected. On the other hand, common “reverse engineering” methods use domain-specific assumptions and are not effective in generic scenarios where the presumed conditions are violated. In this paper, a Backward Mapping Methodology for Design Synthesis (BMDS) is presented that can help conduct design synthesis rapidly and robustly at early stage of parametric engineering design. BMDS is a surrogate model-based approach that combines the strengths of metamodeling and statistics. It can help designers explicitly identify the robust design solutions that satisfy the designer-specified performance requirements through a “backward mapping” from the performance space directly to the design space. Preliminary case studies show its effectiveness and potential to be used as a generic early stage parametric design synthesis methodology in the future.