For a case study of Typhoon Ewiniar performed with a mesoscale model, we compare stratospheric gravity wave (GW) momentum flux determined from temperature variances by applying GW polarization relations and by assuming upward propagating waves, with GW momentum flux calculated from model winds which is considered as a reference. The temperature-based momentum-flux profile exhibits positive biases relative to the reference, which fluctuate significantly with altitude. The vertically-averaged magnitude of the positive biases is about 14% of the reference momentum flux. We found that this deviation from the reference stems from the interference between upward and downward propagating waves. The downward propagating GWs are due mainly to partial reflections of upward propagating waves at altitudes where the background wind and stability change with height. When the upward and downward propagating waves are decomposed and their momentum fluxes are calculated separately from temperature perturbations, the fraction of the momentum flux arising from the downward propagating waves is about 4.5–8.2% of that from the upward propagating waves. The net momentum flux of upward and downward propagating GWs agrees well with the reference from the model wind perturbations. The implications of this study for the GW momentum-flux observations from satellites are discussed.