In this paper we present our current approach to the problem of defining and detecting anomalous tilt behavior. To establish what is considered to be normal tilt behavior, we isolate systematic signals such as hydrologic, thermal, tidal, cultural, and equipment-related effects from the tilt data. The kinds of tilt signals which remain after rejection of the systematic signals are designated by ourselves as residual tilt. Residual tilt consists of asystematic random noise and anomalous tilts. To affirm or deny the contention that an anomalous tilt is present in the data requires the formulation of a statistically valid judgment criteria. Our approach adopts the hypothesis that the random walk model is not significantly different from the residual tilt and allows the application of standard statistical tests to the problem of detecting anomalous varia ions in random noise. In our study of the data analyzed so far, we find that the boundary for detectability is inverse frequency dependent, and this limits the way in which anomalies can be treated. The fact that the magnitude of the anomaly decreases as the tilt data span increases suggests that further criterion development is necessary and tends to imply that longer anomalies will not be detected unless there is a correspondingly larger amplitude. From our studies of three earthquake-association anomalies this does not appear to be the case.