Sensory computations evolve as a function of internal state: are we hungry or satiated, sleepy or alert? To investigate how the brain implements this flexible processing of sensory information, we chose to study the larval zebrafish because of its well-defined behavioral repertoire and its optical accessibility for brain imaging during natural behavior. In this study, we recorded brain-wide activity at single cell resolution in freely moving fish while simultaneously presenting whole-field, dark flash visual stimuli. Behaviorally, dark flashes elicited high amplitude turn responses in awake, alert fish. Notably, these responses were abolished during quiescence. Neurally, dark flashes drove robust visual responses in single neurons spanning multiple brain regions. Despite the behavioral variability, neural responses to the stimulus remained stable across waking and quiescent states. However, we found that the inferred functional connectivity between dark flash responsive neurons and their downstream, motor neuron targets was suppressed by quiescence. This led us to hypothesize that changes in inter-brain region connectivity might underlie the gating of behavioral responses. To investigate this further, we used Reduced Rank Regression to identify the dimension of visual neuron activity that was most predictive of activity in downstream motor neurons. We discovered that the weights defining this dimension were highly dependent on internal state. Specifically, the projection of dark flash evoked activity onto these weights changed significantly between quiescent and non-quiescent trials, despite the visually evoked response being stable. Thus, state-dependent changes in connectivity weights between brain regions may underlie sensorimotor gating of an innate, natural behavior in the zebrafish.