BACKGROUND: This pilot study presents a method of rehabilitation based on the assumption that there is a widely distributed neural network that supports semantic processing and that can be used to improve word retrieval without explicit naming. It employs functional imaging to help understand the neural basis of response to treatment. AIMS: There are both behavioural and neural questions. First, this study investigates whether intensive therapy designed to reduce production errors by eliminating oral responses can yield improved naming of trained items. Second, it investigates whether changes in naming will be associated with changes in fMRI activation and whether the changes in activation pattern associated with recovery are primarily in the right hemisphere (RH) or in perilesional tissue. METHODS AND PROCEDURES: Four weeks of intensive semantic therapy designed to activate the semantic network and minimise errors was administered to AT, an individual with Wernicke's aphasia. Pre- and post-therapy behavioural testing included standard tests of aphasia, naming of pictures from training sets, and a narrative speech sample. Functional imaging of experimental and control tasks was completed pre- and post-therapy. OUTCOMES AND RESULTS: After 4 weeks of training, AT demonstrated improved naming and showed increased use of nouns in narrative speech. FMRI demonstrated increased activation of the left inferior frontal cortex during verb generation as well as increased inferior posterior temporal RH activation. CONCLUSIONS: These results support the effectiveness of intensive intervention methods that require semantic judgements rather than naming and the use of fMRI to understand the neural basis of the response.