Source retrieval is a part of plagiarism discovery process, where only a selected set of candidate documents are retrieved from a large corpus of potential source documents and passed for detailed document comparison in order to highlight potential plagiarism. This paper describes used methodology and the architecture of source retrieval system developed for PAN 2015 lab on uncovering plagiarism, authorship, and social software misuse. The system is based on our previous systems used at PAN since 2012. The majority of features were adopted with some improvements described in this paper. The paper analyzes used methodology and discuss the queries performance. The paper provides explanation for many implementation settings in the source retrieval process. The source retrieval subsystem forms an integral part of a modern system for plagiarism discovery.