We propose Discovering Novel Anomalous Patterns (DAP), a new method for continual and automated discovery of anomalous patterns in general datasets. Currently, general methods for anomalous pattern detection attempt to identify data patterns that are unexpected as compared to “normal” system behavior. We propose a novel approach for discovering data patterns that are unexpected given a profile of previously known, both normal and abnormal, system behavior. This enables the DAP algorithm to identify previously unknown data patterns, add these newly discovered patterns to the profile of “known” system behavior, and continue to discover novel (unknown) patterns. We evaluate the performance of DAP in two domains of computer system intrusion detection (network intrusion detection and masquerade detection), demonstrating that DAP can successfully discover and characterize relevant patterns for these two tasks. As compared to the current state of the art, DAP provides a substantially improved ability to discover novel patterns in massive multivariate datasets.