Abstract: Due to its excellent physical and chemical properties, obsidian was widely used in prehistoric tool production and exchange. Identifying its provenance is crucial for understanding mechanisms of resource acquisition and cultural interaction. Traditional provenance approaches primarily rely on unimodal data, such as elemental composition or Raman spectra, which are often insufficient for distinguishing samples with overlapping compositions or similar structures. This study proposes a two-dimensional chemical fingerprinting approach that integrates elemental and Raman spectral features. Multi-modal data were acquired using X-ray fluorescence (XRF), laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS), and Raman spectroscopy, and classification models were constructed using principal component analysis (PCA). Using obsidian samples unearthed from three archaeological sites (A, B, and C) in the Tibet Plateau and Southwest China as the research objects, the results demonstrate that unimodal models exhibit significant biases in sample classification, whereas the two-dimensional chemical fingerprint model achieves clear discrimination among the three sample groups, thereby markedly improving provenance accuracy and model robustness. This study provides effective technical support for deciphering prehistoric raw material circulation and cultural transmission pathways, offering substantial theoretical significance and promising application prospects.