<p>This paper investigates optimal design strategies for lack-of-fit (LOF) tests in multifactor regression experiments, where the assumed linear model may be misspecified due to the presence of higher-order terms in the true response surface. To address this challenge, we develop a design criterion based on the <i>maximin</i> principle, which provides robustness by balancing parameter estimation efficiency with sensitivity to model inadequacy. Within the framework of response surface methodology, we derive the theoretical properties of the criterion and construct optimal designs under various model assumptions. Worked examples and simulations demonstrate that the proposed maximin designs maintain Type&#xa0;I error control at the nominal level while achieving substantially higher LOF power than classical central composite designs, particularly at moderate curvature levels. The methodology is further supported by a user-friendly algorithm and reproducible <Emphasis FontCategory="SansSerif">R</Emphasis> code, enabling practitioners to readily generate designs for arbitrary (<i>n</i>,&#xa0;<i>k</i>,&#xa0;<i>r</i>) configurations. The results underscore the practical utility of the proposed approach and point toward promising directions for extending the methodology to higher-order alternatives and more complex regression settings.</p>