In software systems, bug localization is a key step in the bug fix process. By automatically narrowing down potential bug locations, the difficulty of bug fix is greatly reduced. In this paper, a bug localization method based on Gaussian processes, called Gaussian processes bug localization (GPBL) is proposed. This method can facilitate fixing bugs for the developers, by recommending source files that may contain bugs. In order to evaluate GPBL, the open-source software Eclipse and Argouml are employed as data sources. Experimental results show that GPBL can achieve 87.16% recall and 78.90% precision on average. In addition, GPBL can locate relevant buggy files more accurately compared with LDA-based bug localization methods.