Techniques for detecting and classifying a posture state of a patient are disclosed, wherein a posture state includes at least one of a posture and/or an activity state related to motion. In one embodiment, one or more signals indicative of at least one of posture and activity state of the patient may be sensed by a sensor. Control logic may be provided to process the one or more sensed signals in a selectable manner. This selectable manner may be based on current posture state data describing a posture state in which the patient has previously been classified. Alternatively or additionally, this current posture state data may describe a posture state transition previously undergone by the patient. The one or more signals that are so processed may then be used to classify a posture state of the patient.