Digital terrain model (DTM) has played an important role in 3D designing, visual analysis and 3D geological modeling in large-scale hydropower engineering. As the pivotal base of 3D visualization and modeling, DTM should be characterized by high precision, less storage and well interactivity during graphic operation. Considering the diversity of data source and taking advantage of two data structures, triangulated irregular network (TIN) and non-uniform rational B-splines (NURBS), a novel methodology is presented for reconstructing engineering terrain of hydropower project. With integration of multi-source data, enhanced Delaunay algorithm is introduced to rebuild the TI NDTM, which is a terrain surface in TIN and a faithful depiction of complex topography but in low-memory efficiency. Based on the TIN model, applying section scanning sampling and linear interpolation, the transformation from discrete, irregular and diverse data to continuous and regular sampling cross-sectional curve sequence, is realized. The appropriate compression of the sampling data is also imposed to be performed for guaranteeing the following reconstruction work. Eventually, employing the NURBS technique and skinning method, the NURBS-DTM, which represents a NURBS surface and satisfies the requirement after precision assess with weighted errors, is reconstructed with the intermediate data. Meanwhile, there is another achievement that two databases of terrain data, one from initial data and the other from sampling data, are established for repeatable reconstruction with different demands. With the successful application of the presented method, a stable foundation is laid for 3D engineering geological modeling, visual designing and analysis of the hydropower projects.