Artistic text recognition poses significant challenges due to the diversity of fonts, visual effects, and background noise—elements designed for aesthetics yet detrimental to accurate recognition. While the skeletal structure of characters remains stable across variations, existing methods have underutilized this information for robust recognition. In this work, we propose Skeleton-Guided Artistic Text Recognition (SG-ATR), a novel method that explicitly leverages skeletal representations to enhance text feature extraction. Our framework integrates a Skeletonization module to distill essential character structures while mitigating interference from complex backgrounds. To advance benchmarking in artistic text recognition, we introduce Artistic Text-In-The-Wild (ATTW), a dataset comprising 16,627 diverse artistic text instances, curated from widely recognized sources and spanning a broad spectrum of artistic styles. Extensive experiments demonstrate that SG-ATR achieves 75.39% % accuracy on ATTW, surpassing state-of-the-art (SOTA) methods. Furthermore, SG-ATR generalizes well to standard scene text recognition benchmarks, underscoring its robustness beyond artistic text scenarios.