When scholars want to make use of the best mechanisms for reference extraction, this endeavour invariably involves training the available tools for their particular dataset (and, eventually, for the particular task of reference extraction, too). Unfortunately, training routines, and - even more cumbersome - data formats to use for such routines differ considerably between the most promising tools. In order to provide some orientation for both scholars planning a reference extraction project, and those considering offering some data they are in possession of as training data, this presentation will discuss (a) the most important data formats and (b) data preparation workflows for such training data. In the former part, besides the data formats themselves, some conversion mechanisms are mentioned. In the latter part, various tooling options for the manual or semi-manual tagging of references in source texts will be presented briefly.