Accumulating sensory information over long-time scales is a critical component of most circuit computations in the brain across species. It remains poorly understood how such dynamics are implemented mechanistically at the synaptic level. Dissecting such neural circuits in vertebrates requires precise knowledge of functional neural properties and correlating dynamics with the underlying wiring diagram within the same individual animal. Here we combine functional calcium imaging with ultrastructural circuit reconstruction, using a visual motion accumulation task in larval zebrafish. Using connectomic reconstructions of functionally identified cells and modeling we show that bilateral and interhemispheric inhibition as well recurrent connectivity within anterior hindbrain circuitry are essential for sensory accumulation and motor decision making. We also show that sensory denoising is implemented by an ipsilateral dis-inhibitory input provided by a small population of tonically actived direction selective neurons. Lastly, we demonstrate that details of this circuitry can be obtained through targeted fluorescent photo-conversion labeling of functionally identified neuronal response types paired with a whole brain connectomics resource. This enables statistical connectome analysis, sensory inputs and motor outputs identification and cross-sample comparison. This general approach combining light and electron microscopy allows to constrain models of neuronal computations in the vertebrate brain.