Rubber composite formulation design requires the simultaneous optimization of multiple properties, whereas orthogonal experiments are often limited by small sample sizes and responses with different dimensions and optimization directions. To address this issue, this study proposes a multi-objective performance optimization framework based on grey relational analysis and orthogonal experiments. First, the multi-response orthogonal experiment optimization problem is formally defined, and an integrated procedure involving response normalization, reference sequence construction, grey relational analysis, and signal-to-noise ratio optimization is established. Second, equal-weight and weighted grey relational models are developed, with their applicability examined through weight sensitivity analysis and index permutation invariance. The results indicate that point-to-point grey relational methods are more suitable for evaluating parallel performance indicators. The proposed framework can identify promising formulation regions within a limited experimental space and provide interpretable guidance for key-variable selection and subsequent extrapolation, making it suitable for small-sample, multi-factor, and multi-objective optimization of rubber composites.