Injection of liquid waste into a highly transmissive, saltwater-bearing, fractured dolomite underlying the city of St. Petersburg, Florida, provided an opportunity to study density-dependent flow associated with two miscible and density-different liquids. The injection zone was 98 m thick with a radial hydraulic conductivity of 762 m/d and a vertical hydraulic conductivity of 152 m/d. Mean chloride concentrations of the injectant during two tests of 91 and 366 days duration were 180 and 170 mg/L, respectively, whereas chloride concentration of native salt water ranged from 19,000 to 20,000 mg/L. During the 366-day test, chloride concentration in water from a well open to the upper part of the injection zone 223 m from the injection well approximately stabilized at about 4000 mg/L. Relatively constant chloride concentrations in water from this observation well at a level significantly greater than the injectant concentration suggested the hypothesis that circular convection with saltwater flow added chloride ions to the injection zone flow sampled at the observation well. In order to assess the acceptability of the circular convection hypothesis, information was required about the velocity field during injection. Mass transport model simulations were used to provide this information, after determining that the fractured injection zone could be treated as an equivalent porous medium with a single porosity. The mass transport model was calibrated using the 91-day test data from two observation wells 223 m from the injection well. The model was then run without parameter changes to simulate the 366-day test. Mass fractions of injectant computed for four observation wells during the 366-day test compared favorably with observed mass fractions. Observed mass fractions were calculated as a function of chloride concentration and density. Comparisons between model-computed mass fraction and velocity fields in a radial section showed circular convection, with salt water flowing toward the injection well in the lower part of the injection zone. The salt water then mixed with the injectant, and the mixture flowed away from the injection well in the upper part of the injection zone. On the basis of the model results and the assumed reasonableness of treating the injection zone as an equivalent porous medium with a single porosity, the hypothesis of circular convection with saltwater flow during subsurface injection of liquid waste into a highly transmissive saltwater-bearing fractured dolomite was judged acceptable.