Résumé : We report a single case study of a brain-damaged patient, ER, who showed a remarkably consistent category-specific deficit for living things. His impairment was observed across tasks (naming, definition, matching, drawing from memory, questionnaires), input modalities (visual, verbal, nonverbal auditory), and output modalities (verbal vs. pointing or visual matching responses) as well as for different types of knowledge. Although visual knowledge of living things was severely affected, his category-specific impairment in nonverbal sound recognition is inconsistent with models of category-specific deficits based on pre-semantic visual descriptions. ER's deficit cannot fully be explained by item typicality, word frequency, visual complexity, homomorphy, age of acquisition, value to perceiver, or modality of transaction. Furthermore, in ER, contextual cues were even slightly detrimental for the recognition of animals. ER's naming and recognition errors were constrained by the categorical structure of the knowledge base: In most cases they respected both the second- and first-order superordinates. In particular, ER's knowledge of shared categorical properties related to biological function was almost spared. This result is compatible with the idea that, for living things, shared functional properties and shared perceptual properties are strongly correlated. Feature-based models assuming perceptual vs. functional semantic components cannot account for ER's deficit, since for living things he was impaired on both kinds of features to a similar extent. ER's behaviour is quite consistent with the notion that conceptual knowledge is organised categorically in the brain, with one or several specialised subsystems for biologically related entities.