Recurrent connections are a fundamental feature of neural circuits, enabling feedback loops that regulate network activity. Connectome datasets show strong recurrent connections in the fruit fly visual system, where motion information is processed. However, current models of motion detection focus on feedforward connectivity, leaving the role of recurrence largely unexplored. Here, we investigate the impact of recurrent connectivity on “ON” (bright stimuli) motion selectivity using connectome-constrained deep mechanistic networks (DMNs) of the fly visual system. DMNs feature a one-to-one correspondence between every neuron and synaptic connection in the fly and their model counterparts, enabling detailed predictions of neural function. To capture the nonlinear neural computations involved in motion detection, we incorporate biologically realistic conductance-based synapse dynamics, which accurately predict known single-neuron tuning properties. Using our DMN model, we predict the computational role of recurrent connectivity within the inputs to T4 cells, the first motion-detecting neurons in the ON pathway. Highlighting the predictive power of our model, we first demonstrate that simulated ablations of major T4 input cell types reproduce the known decrease in T4 motion tuning. We then examine the unknown role of recurrent inhibition between Mi4 and Mi9 cells, the primary inhibitory inputs to T4, by ablating these connections in silico. Notably, these ablations reduce T4 direction selectivity, thus weakening responses to preferred direction motion and strengthening responses to non-preferred directions. This suggests that the Mi4-Mi9 recurrent connectivity sharpens T4 motion detection. Studying the underlying mechanism, we find that the feedback inhibition from Mi4 to Mi9 fine-tunes both the timing and amplitude of input signals to T4, enhancing its direction selectivity. In summary, our results predict that recurrent connectivity, mediated by Mi4 and Mi9 cells, plays a crucial role in enhancing motion detection in the fly visual system by synchronizing and modulating the inputs to T4 cells.