Cognition may reasonably be distinguished into world-estimation and planning tasks. Our focus in this work is the world estimation task, i.e., the task of establishing and updating beliefs about the world. One aspect of being intelligent in this task is noticing one’s mistakes and correcting them. An intelligent system may be realized by dividing its operation into a base-level reasoning system and a metareasoning system. The base-level system is responsible for processing inputs from the world and recording its conclusions in a belief state. The metareasoning system monitors the base-level system so that it can detect symptoms of errors in the belief state and attempt belief revisions. We describe and evaluate such a system in this report. The base-level system is an abductive reasoner responsible for finding explanations for inputs given as reports of putative observations. When no plausible, consistent explanation is forthcoming for some reports, we say these unexplainable reports are anomalous. The presence of anomalies is a symptom of errors in the belief state since, in the usual case, all reports should be explainable. However, sometimes the anomalous reports are not reports of observations but rather false or noisy reports. An abductive metareasoning system attempts to explain anomalies as errors of various kinds and makes the appropriate revisions. If a sufficiently plausible explanation is not found, an anomalous report is attributed to noise. We evaluate this two-level system in a pair of object tracking tasks, one simulated and one based on aerial surveillance. Both tasks are challenging due to limited sensor capabilities and a very high level of noise. The proposed two-level system realizes significantly more accurate belief states and better noise detection than a system that lacks abductive metareasoning.