Experimental measurements of the solubility of CO2, CO, CH4, H2, and H2O in silicate melt were made at pressures from 10 to 20 Kb and temperatures of 1200¿C. The C-O-H fluid phase was buffered by graphite, iron, and w¿stite. Detailed measurements were made on melts of NaAlSi3O8 composition and the results confirmed for alkalic and tholeiitic basalt liquids. The melt volatiles have dramatically lower H2/H2O and higher CO/CO2 ratios than the fluid. The result is large differences between the C-O-H compositions of the melt and the fluid (fluids about 16 atm % C, 6% O, and 78% H; melts about 4% C, 32% O, and 64% H). The measured solubilities allow calculation of activity coefficients for volatile species in the melt. They are, roughly, CO=0.1, H2O=0.6, CO2=1, H2=3, and CH4=60. This two-order-of-magnitude variation in activity coefficients causes the significant volatile fractionation observed between fluid and melt. We argue that the same fractionation will occur in the more usual case of fluid-absent partial melting of planetary mantles. The fractionation is such that reducing mantles will form magmas that yield volcanic gases about 1.5 orders-of-magnitude more oxidizing than the mantle source, while oxidizing mantles will form magmas with more reduced volcanic gases. We conclude that a volcanic gas composition cannot be directly used to estimate either the oxygen fugacity or volatile composition of its magma source region. The results suggest that volcanic gases will normally lie in the ''neutral'' range of oxygen fugacity, near that of the quartz-fayalite-magnetite buffer. These gases are predominately H2O with minor CO2, CO, H2, and CH4.