Geosocial networks such as Foursquare have access to users' location information, friendships, and other potentially privacy sensitive information. In this paper, we show that an attacker with access to a naively-anonymized geosocial network dataset can breach users' privacy by considering location patterns of the target users. We study the problem of anonymizing such a dataset in order to avoid re-identification of a user based on her or her friends' location information. We introduce k-anonymity-based properties for geosocial network datasets, propose appropriate data models and algorithms, and evaluate our approach on both synthetic and real-world datasets.