Construction of 3D geological models is a very efficient way to create a good understanding of geological features. It is a comprehensive method to geologists, geophysics engineers and GIS experts who sometimes need to visualize an area to accomplish their researches. Geological drill hole records have been widely used as the main data sources in 3D geological model construction approaches. However, since the original drill hole sampling points are limited and maldistributed, it is difficult to construct 3D geological models that not only show favorable visual effect but also reflect accurately the geological structure of surveying regions. To solve this problem, this paper proposes a 3D geological model construction approach based on virtual boreholes. Firstly, according to the spatial distribution of original drill holes, positions of interpolated points are determined self-adaptively and Kriging interpolation algorithm is applied to the construction of interpolated virtual boreholes. Secondly, according to the pinch-out rules, the pinch-out virtual boreholes are constructed, and the improved self-adaptive butterfly subdivision algorithm is applied to the construction of subdivided smooth virtual boreholes. Finally, with the original drill holes, the interpolated virtual boreholes, the pinch-out virtual boreholes and the subdivided smooth virtual boreholes, 3D geological models of survey regions can be constructed, matching geostatistics patterns, being C¹ geometric continued, and showing favorable visual effect. Differing from the traditional modeling methods that regard virtual boreholes as complements to expert knowledge, the proposed method introduces virtual boreholes into each step of the construction of three-dimensional geologic models as the intermediate elements of the core algorithm, simplifying the realization process of the modeling algorithm effectively, and ensuring the stability and efficiency of the algorithm.