Using dense seismic data sets, we present a new 3D velocity model of Taiwan that images aprominent mid-crustal (˜20-30 km) high-velocity body beneath west-central Taiwan. The inclusion of high-quality post-2012 recordings from the Central Weather Administration Seismic Network (CWASN) ensures theelimination of uncorrectable timing errors. In addition, a machine learning-based phase picker was applied tothe entire data set to improve the consistency and accuracy of phase arrival identification. The resulting modelgenerally aligns with previous tomographic studies. Structures inferred from velocity gradients in thetomographic profiles largely correspond to mapped faults and geological unit boundaries in Taiwan. The modelalso images a pronounced mid-crustal (˜20-30 km) high-velocity anomaly beneath west-central Taiwan. Underrepresentative P-T conditions, Vp-Vs-density comparisons indicate the best match with mafic compositions,though the interpretation is not unique. This mafic interpretation is compatible with passive-margin maficadditions (underplating and/or intrusions). The anomaly coincides with reduced seismicity below ˜20 km anddepth-dependent stress orientations, consistent with a relatively competent mid-crustal volume. Geological andgeophysical similarities with the Dongsha Rise further suggest a possible shared tectonic and magmatic origin,likely linked to mafic underplating during South China Sea rifting. These findings improve our understanding ofstructural highs along passive continental margins and their role in influencing crustal deformation in theTaiwan orogen. The new model also provides a robust framework for future waveform-based seismic imaging.