We used positron emission tomography to investigate brain activity in response to hearing or reading nouns of varying imageability. Three experiments were performed. Activity increased with noun imageability in the left mid-fusiform gyrus, the lateral parahippocampal area in humans, and in the rostral medial temporal lobes close to or within perirhinal cortex. The left mid-fusiform activation has been observed in previous imaging studies of single word processing. Its functional significance was variously attributed to semantic processing, visual imagery, encoding episodic memories, or the integration of lexical inputs from different sensory modalities. These hypotheses are not mutually exclusive. The more rostral medial lobe response to noun imageability has not been observed previously. However, lesions in perirhinal cortex impair knowledge about objects in non-human primates, and bilateral rostral ventromedial temporal lobe potentials in response to object nouns were observed with human intracranial recordings. Imageable (object) nouns are learnt with reference to sensory experiences of living and non-living objects, whereas acquisition of the meaning of low imageable (abstract) nouns is more dependent on their context within sentences. Parahippocampal and perirhinal cortices are reciprocally connected with, respectively, second and third order sensory association cortices. We conclude that access to the representations of word meaning is dependent on heteromodal temporal lobe cortex, and that during the acquisition of object nouns one route is established through ventromedial temporal cortical regions that have reciprocal connections with all sensory association cortices.