Summary Single-phase upscaling techniques have been in use for a number of years now to incorporate subgrid reservoir description information into reservoir simulation. There have also been a number of approaches to developing two-phase upscaling algorithms suitable for black oil simulation. The key barrier to the development of compositional upscaling techniques is algorithm speed. A straightforward extension of existing two-phase upscaling approaches would consume enormous amounts of computer time because of the number of additional components modeled and the need to perform flash calculations. This paper describes an upscaling method for simulations in which compositional processes play a key role, such as miscible gas injection or gas cycling schemes. We will describe the method to generate upscaled compositional fluxes in two or three dimensions, and the way in which the upscaled fluxes are incorporated into a compositional simulator. Our approach uses a streamline technique, which provides a tremendous gain in speed for the loss of a small amount of accuracy. The streamline technique produces upscaled fluxes almost identical to those obtained from postprocessing conventional compositional model runs in a fraction of the time. The results of applying these upscaled fluxes in a realistic example show that we can obtain close agreement between fine and coarse grid models of lean gas injection into oil at residual saturation. For an example problem with a 12-component equation of state (EOS), the component recovery curves with upscaled fluxes on a 5 × 5 grid agree well with the original 100 × 20 fine grid