Surface defect detection for workpieces with inner holes is a critical task in manufacturing quality control. However, current methods often struggle to satisfy industrial standards due to the scarcity of defective samples and the subtle, indistinct nature of surface anomalies. To address these challenges, we propose a novel anomaly detection framework based on Split-Attention Patch Distribution Modeling. By integrating the split-attention mechanism, our approach significantly enhances feature extraction capabilities, enabling more robust discrimination between normal and anomalous patterns even amidst complex backgrounds. Specifically, we employ multivariate Gaussian distribution modeling to accurately capture intra-feature correlations and utilize the Mahalanobis distance for precise anomaly scoring. Extensive experiments demonstrate that our model outperforms state-of-the-art algorithms in both AUROC and precision. Notably, the proposed method exhibits exceptional robustness in the challenging task of tiny defect detection, marking a significant improvement over existing baselines.