Résumé : Publishing research data aims to improve the transparency of research results and facilitate the reuse of datasets. In both cases, referencing the datasets that were used is recommended. Research data repositories can support data referencing through various measures and also benefit from it, for example using this information to demonstrate their impact. However, the literature shows that the practice of formally citing research data is not widespread, data metrics are not yet established, and effective incentive structures are lacking. This article examines how often and in what form datasets published via the research data repository RADAR are referenced. For this purpose, the data sources Google Scholar, DataCite Event Data and the Data Citation Corpus were analyzed. The analysis shows that 27.9 % of the datasets in the repository were referenced at least once. 21.4 % of these references were (also) present in the reference lists and are therefore considered data citations. Datasets were referenced often in data availability statements. A comparison of the three data sources showed that there was little overlap in the coverage of references. In most cases (75.8 %), data and referencing objects were published in the same year. Two definition approaches were considered to investigate data reuse. 118 RADAR datasets were referenced more than once. Only 21 references had no overlaps in the authorship information -- these datasets were referenced by researchers that were not involved in data collection.