Ray theory has been used to attempt to explain the propagation of very low-frequency (1–20 Hz) noise generated by downwind wind turbines. Two NASA field experiments with a large downwind machine at Medicine Bow, Wyoming, show that the downwind sound decays by spherical spreading near the source, but by cylindrical spreading downrange. Ray theory calculations explain this behavior. The favorable wind gradient downwind causes rays to refract downward and bounce along the ground. A sound channel is built up by repeated jumps in the number of ray arrivals at the receiver as downwind range is increased. Near the source, where no multiply reflected rays arrive, the sound field spreads spherically. The first jump (onset of multiple arrivals) is predicted to occur at about 2 km for the conditions of the NASA experiments. The second NASA experiment confirms this prediction. For the upwind sound, a shadow zone is predicted because of the unfavorable wind gradient. The NASA measurements show only that the upwind SPL in the predicted shadow zone is of order 7–15 dB less than the downwind SPL at comparable ranges. [Work supported by NASA.]