Positron Emission Tomography (PET) with the tracer H215O was used to measure regional cerebral blood flow in 13 healthy volunteers during two experimental memory tasks, one of which was well-practiced and the other of which was novel. The materials used for the memory tasks consisted of two complex narratives (Story A and Story B from the Wechsler Memory Scale). Natural language materials were chosen because they similate experimentally the natural learning situation and permit study of the neural mechanisms by which recall memory becomes more fluid, automatic, or "rote." One week before the PET study, subjects were trained to perfect recall of Story A, while they were exposed to Story B only 60 s prior to PET data acquisition. Despite the substantial differences in level of familiarity (and in free recall performance), patterns of activation were quite similar; activations presumed to reflect recall in both tasks included frontal, inferior temporal, thalamic, anterior cingulate, and cerebellar regions. Many regions were smaller during recall of the familiar story, however, presumably reflecting greater neural efficiency due to practice. In addition, the novel task activated an additional left frontal region that is presumed to reflect more active encoding. The similarity and multiplicity of the activations in the two tasks suggest that the brain uses a multinodal general network for memory tasks such as free recall, while the differences suggest that some nodes in the network may be used for specific components of memory such as encoding and retrieval.