This paper shows how Social Network Analysis, the study of relational data in specific social environments, can be used to summarize multiparty radio news recordings. A social network is extracted from each recording and it is analyzed in order to detect the role of each speaker (e.g. anchorman, guest, etc.). The role is then used as a criterion to select the segments that are more representative of the recording content. The results show that the length of the recordings can be reduced by more than 90 percent while still preserving most of the information about their content.