Restricted semantic fields and resultant stimulus overselectivity are often thought to be typical of low-functioning autism, as is a strong visual processing preference. However, these conclusions may in part be an artifact of testing methodology. A 12-year-old, low-functioning and nonverbal autistic boy was tested on an auditory word-to-picture selection task. The picture foils were chosen to have visual features, semantic features, both, or neither in common with the correct answer. Errors were made more often to semantically than to visually related items, and he showed generalization to items that had not been explicitly trained. This is taken as evidence that his semantic fields are broader than otherwise apparent, and that he was capable of expanding his semantic representations independently of specific training.