It is an old problem in rock physics that the saturation dependence of high-frequency laboratory velocities does not match the Biot-Gassmann theory commonly used to predict the effects of gas on seismic velocities. A new interpretation of laboratory velocity data shows that the saturation dependence is controlled by two previously published high-frequency acoustic mechanisms: (1) a gas pocket model that describes pressure equilibration between liquid and gas-saturated regions of the pore space, and (2) local fluid flow, induced by pressure equilibration in pores with different aspect ratios. When these two mechanisms are added to Biot theory, the result describes published velocity versus gas saturation data for a wide range of rock types. These two mechanisms are negligible at the lower frequencies of seismic data, so the saturation dependence of laboratory velocities cannot be used to predict the saturation dependence at seismic frequencies. The one laboratory measurement that is relevant for predicting the seismic velocity is the ultrasonic velocity of the dry rock. The dry-rock velocities should be used in the Biot-Gassmann theory to predict the full saturation dependence of the seismic velocities.