The modular multilevel converter (MMC) has become one of the most promising converter topologies for medium/high-power applications. As the MMC is structured based upon stacking up of a number of series-connected identical SubModules (SMs), to improve its fault tolerance and reliability, SM failure detection and location is of significant importance. In this paper, two SM-failure detection and location methods are proposed, i.e., a clustering algorithm (CA)-based method and a calculated capacitance (CC)-based method. In the proposed CA-based method, a pattern recognition-based fault diagnosis approach is developed, which employs the clustering algorithm to detect and locate the faulty SMs with open-switch failures through identifying the pattern of two-dimensional trajectories of the SM characteristic variables. The proposed CC-based method is based on calculation and comparison of a physical component parameter, i.e., the nominal SM capacitance, and is capable of failure detection, location, and classification within one stage. Performance of the proposed failure detection methods for an MMC system is evaluated based on time-domain simulation studies in the PSCAD/EMTDC software environment. The reported study results demonstrate the capabilities of the two proposed methods in detecting and locating any SM failure under various conditions accurately and efficiently.