User authentication based on fingerprints is vulnerable to dictionary attacks. Recent research has demonstrated the possibility of generating a small number of 'MasterPrints' that can fortuitously match the fingerprints of a large number of identities. The problem is particularly exacerbated for partial prints such as those used in smartphones. Such systems often store multiple templates per user (e.g., multiple impressions of a single finger) to compensate for the limited size of the sensor, variation in finger placement, and other types of intra-class variations. The presence of multiple templates, however, increases their chances of matching against a MasterPrint thereby compromising security. This paper presents a novel technique to perform template selection in such a way that the chance of MasterPrint attack gets reduced. Experiments conducted using a commercial fingerprint matcher on two datasets indicate that the proposed approach can be effective against MasterPrint attacks whilst retaining verification performance.