An animal's survival depends on its ability to correctly evaluate sensory stimuli and select appropriate behavioral responses. When confronted with ambiguous stimuli, the brain is faced with the task of selecting one action while suppressing others. Although conceptually simple, the site and substrate of this elementary form of decision making is still largely unknown. Zebrafish larvae respond to a moving dot stimulus in either of two ways: a small object (potential prey) evokes approach, whereas a large object (potential predator) is avoided. The classification of object size relies on processing in the optic tectum. We genetically identified a population of cells, largely comprised of glutamatergic tectal interneurons with non-stratified morphologies, that are specifically required for approach toward small objects. When these neurons are ablated, we found that the behavioral response is shifted; small objects now tend to elicit avoidance. Conversely, optogenetic facilitation of neuronal responses with channelrhodopsin (ChR2) enhances approaches to small objects. Calcium imaging in head-fixed larvae shows that a large proportion of these neurons are tuned to small sizes. Their receptive fields are shaped by input from retinal ganglion cells (RGCs) that are selective for prey identity. We propose a model in which valence-based decisions arise, at a fundamental level, from competition between dedicated sensorimotor pathways in the tectum.