A memory anomaly detection method and device are provided. The method includes: obtaining a first memory log, where the first memory log includes log information of a plurality of garbage collections, and log information of each garbage collection includes a garbage collection time, and includes at least one of a downtime, memory usage after garbage collection, and memory usage before garbage collection; obtaining, based on log information in a first detection time window, first statistical information corresponding to the first detection time window; and determining, based on the first statistical information corresponding to the first detection time window, an anomaly degree corresponding to the log information in the first detection time window. It can be learned that by implementing the method described in the embodiments of this application, accuracy of a detection result of a memory anomaly degree is improved.