Abstract. Over the last years, thanks to Open Data initiative and the Semantic Web, there has been a vast increase on user contributed data. In several cases (e.g. OpenStreetMap, Geonames), the respective data include geospatial information, that is the coordinates and/or the precise geometries of buildings, roads, areas, etc. In such cases, proper schemas are defined to allow users to annotate the entities they contribute. However, browsing through a large and unstructured list of categories in order to select the most fitting one might be time consuming for the end users. In this paper, we present an approach for recommending categories for geospatial entities, based on previously annotated entities. Specifically, we define and implement a series of training features in order to represent the geospatial entities and capture their relation with the categories they are annotated with. These features involve spatial and textual properties of the entities. We evaluate two different approaches (SVM and kNN) on several combinations of the defined training features and we demonstrate that the best algorithm (SVM) can provide recommendations with high precision, utilizing the defined features. The aforementioned work is deployed in OSMRec, a plugin for JOSM tool for editing OpenStreetMap.