<p>The accuracy of seismic inversion usually depends on the quality of the initial model. A novel well-log interpolation method based on a pattern-feature correlation (PFC) approach has been developed to build highly accurate initial models for seismic inversion. The interpolation weights in the PFC interpolation scheme are determined based on the correlations of local geological pattern features, which helps construct accurate low-wavenumber initial models. However, the PFC interpolation method lacks stratigraphic constraints and is prone to stratum-crossing of interpolation nodes, which leads to poor accuracy in complex structures. For this purpose, we extract relative geological time (RGT) volume from seismic data using seismic dips estimated by the plane-wave destruction (PWD) method as input. Then, we integrate RGT as constraints in well-log interpolation. Specifically, the interpolation nodes whose time differences from the node to be interpolated exceed a threshold determined approximately as half the thickness of the thinnest layer in the RGT volume are excluded. By this way, extrapolation of well-log data over long distances while preserving the geological consistency is realized. We use this interpolation strategy to build highly accurate <i>P</i>-wave impedance models for poststack seismic inversion. Numerical tests using both synthetic and field data confirm that the initial model constructed with RGT constraints yields an acoustic impedance volume with superior spatial continuity and high fidelity, particularly in complex subsurface structures. The high-quality initial models further enhance the inversion accuracy. Although we only discuss a poststack inversion scheme, the proposed initial model building method can theoretically be applied to prestack inversion methods.</p>