Prepositions can express a variety of different semantic relations, such as time, location, or manner. To automatically disambiguate the meaning of prepositions is a challenging task for natural language processing applications. This work investigates automatic word sense disambiguation and semantic role labeling techniques for prepositions. In particular, it empirically investigates how information from one task can improve performance on the other task and shows that jointly disambiguating word senses and semantic roles is superior to disambiguating them independently.