In this paper, we propose a novel method of automatic detecting urban façade repeatability by using probabilistic reasoning. The main advantage of this algorithm is that all the occluded building components can be detected as well as components that appear in façade image. First, we use the image entropy theory to derive qualified closed regions in image as the candidate repetitive components. Moreover, we introduce a repetitive characteristic surface which can be used to easily determine accurate locations, shapes and sizes of derived components. Finally, we detect all the occluded repetitive components with accurate locations, shapes and sizes by using the Bayesian probability network which obtained through an offline train from a façade images database. Experiment result demonstrates that the proposed algorithm improves the accuracy, robustness and efficiency on façades databases compared with the state-of-the-art methods.