The ultrasonic velocity and absorption of 32 silicate melts have been determined as a function of temperature (1175--1925 K) and frequency 3--12 MHz). The samples include the components SiO2, TiO2, Al2O3, FeO, MgO, CaO, SrO, BaO, Li2O, Na2O, K2O, Rb2O, and Cs2O, and range from simple alkali silicates to multicomponent synthetic and natural melts. The relaxed sound spped varies systematically with subsitutiion of the alkali and alkaline earth oxides, decreasing monotonically with increased cation radius. An algorithm for the compositional dependence of the ultrasonic at 1673 K: 1/c&Sgr;(&khgr;i,v/c¿i,) where &khgr;i,v is the volume fraction of component i and c¿i is independent of composition, is capable of predicting relaxed melt velocities to within 4%. The temperature dependence of the sound speeds in the nondispersive region is quite small, ≈1 part in 104 K-1. A simple linear model for the compositional dependence of the derived property (ΔV/ΔP)T has an uncertainty of 13%, approximately equal to heat capacity. The contributions of the changes in sound speed and density with temperature are nearly the same in their influence on the temperature dependence of the bulk modulus of natural melts. The relaxation properties c/c0 (ratio of the sound speed to the low-frequency value) and αλ (absorption per wavelength) of all the liquids studied are very similar when plotted against ω&tgr;, the product of the angular frequency and the shear relaxation time. Both properties are consistent with &tgr; equal to 1% of the product of the low-frequency shear viscosity and bulk modulus. Dispersion begins when ω&tgr;≂0.1 and the limiting high-frequency sound spped c∞ is about 2.3 times c0. The product αλ attains a maximum value of about 1.4 when ω&tgr;≂0.8. The data cannot be explained by a single relaxation time in the melts; rather, a spectrum of relaxation time is indicated. ¿ American Geophysical Union 1987