Sensory systems dynamically optimize their processing properties in order to process a wide range of environmental and behavioral conditions. However, attempts to infer the function of these systems via modeling often treat system components as having static processing properties. This is particularly evident in the Drosophila motion detection circuit, where the core algorithm for motion detection is still debated, and where inputs to motion detecting neurons remain underdescribed. Using whole-cell patch clamp electrophysiology, we measured the state- and stimulus-dependent filtering properties of inputs to the OFF motion-detecting T5 cell in Drosophila. Simply summing these inputs within the framework of a connectomic-constrained model of the circuit demonstrates that changes in the shape of input temporal filters are sufficient to explain conflicting theories of T5 function. Therefore, with our measurements and our model, we reconcile motion computation with the anatomy of the circuit.