Abstract: Rock mass fractures in dam foundation directly affect structure safety and seepage stability of hydropower projects. Accurate 3-D models of rock mass fracture networks provide a solid basis for analysis of dam safety. In construction of such network models, random simulation methods are convenient in obtaining sample data of fracture parameters from the large quantity and complicated distribution of rock mass fractures, and recent research is focusing on the Monte Carlo method for random sampling. However, such an approach ignores the importance of those fractures lying below the exposed surface layer, and suffers from poor numerical stability, low accuracy, and the problem of sample collapsing. This paper presents a new simulation method of 3-D fracture networks in dam foundation using Latin hypercube sampling (LHS). This method derives fracture data from the photographs of both exposed surface layer and boreholes using camera technologies, then applies LHS to stochastic simulation of fracture parameters including their location, size, quantity, and attitude, and finally constructs a 3-D fracture network model in the frame of VisualGeo. It was applied to a hydropower project in China, along with the Monte Carlo method for comparison. The results show its advantages of lower computational cost and more accurate predictions of the field test values for dam foundation stability analysis.