Build orientation determination is an important pre-processing step in Additive Manufacturing. To identify an optimal build orientation, there are two main tasks, generating a set of alternative orientations and evaluating these alternatives with pre-set criteria. To solve the first task, currently there are two categories of methods, exhaustive computing and continuous surface decomposition. However, for exhaustive computing methods, the infinite original alternative orientation space is an obstacle, especially when considering multiple objectives. While the other type of methods have difficulty on surface separation and shape boundary recognition when facing complex CAD models. To tackle of these obstacles, this paper introduces a new method applying a statistical tool to form facet clusters for decomposing an STL model in a discrete way. The formed facet clusters can be used to efficiently generate meaningful alternative build orientations and can also be used to predict surface quality distribution over a part model for further process planning or design iteration.