Eikonal tomography based on ambient noise data is one of the most effective methods to reveal shallow earth structures. By tracking surface wave phase fronts, constructing travel time surfaces, and computing the gradients of travel time surfaces to generate phase velocity maps, eikonal tomography avoids the ray tracing and matrix construction and inversion in the traditional surface wave tomography methods.In this study, we collect continuous ambient noise data recorded by a dense seismic array in Karamay, Xinjiang to construct a 3D model of shallow structures using eikonal tomography. The seismic array consists of 35 stations with shortest interstation distance close to 1km. 890 empirical surface wave Green's functions (EGFs) between each station pair are retrieved by cross-correlating one or two months of continuous ambient noise data. From these EGFs, surface wave travel times in the frequency range of 1.8 to 4.0 Hz are measured by a frequency–time analysis technique (FTAN). Then, eikonal tomography is adopted to construct Rayleigh wave phase velocity maps and estimate the phase velocity uncertainties. Finally, we invert the obtained phase velocity dispersion curves for 1D shear velocity profiles and then assemble these 1D profiles to construct a 3D shear velocity model. Major velocity features of our 3D model are correlated well with the known geological features. A shallow east–west velocity discontinuity is observed, which clearly reflects the lithological change between Baogutu formation (C1b) and Xibeikulasi formation (C1x) of lower Carboniferous system. Low shear velocities are observed beneath the location of porphyry copper deposit (V), possibly related to stockwork fracture and hydrothermal brecciation developed during the intrusion of deep magma in forming the deposit.