Support Vector Regression often faces challenges in small-sample scenarios due to noise sensitivity and inconsistent error metrics, limiting robustness and generalization. This study proposes an L∞-norm-based regression model, L∞-SVR, along with a primal-dual interior-point optimization algorithm. The approach minimizes the maximum regression error, improving noise resistance and reducing solution space dimensionality for better computational efficiency. Experiments on synthetic and real-world datasets demonstrate that L∞-SVR outperforms existing methods, achieving an average R2 improvement of 1.16 across five real-world datasets, particularly excelling in small-sample and noisy environments.