The present simulation study elaborates on a finite element (FE) computational fluid dynamics (CFD) model (Gerogiorgis and Ydstie, 2003a) developed for a candidate carbothermic aluminium reactor (Johansen and Aune, 2002), aimed at industrial implementation of carbothermic Al production. Carbothermic reduction is an alternative to the conventional Hall-Heroult electrolysis process and is characterized by cost and environmental advantages as well as by a challenging complexity. Process technology encompasses a wide spectrum of phenomena (convection, diffusion, reaction, evapouration, electric field) that occur simultaneously in a multiphase configuration, the geometry of which is an open design problem and remains to be determined without prior experience or even abundance of experimental data. The strong interaction among Joule heating, endothermic reaction, natural Boussinesq convection and turbulent flow phenomena is of paramount importance for understanding reactor performance; conducting CFD simulations is an efficient way to advance with the latter goal, since reliable high-temperature measurements of state variables are remarkably laborious, uncertain and expensive. The quadruple partial differential equation (PDE) problem (electric charge, heat, momentum and gas volume balances) for the slag flow in the ARP reactor is solved via a commercial CFD software suite (FEMLAB (R) v. 2.3) to obtain potential, temperature, velocity and gas volume fraction distributions in a two-dimensional domain, representing in detail the complete second stage of the proposed carbothermic reactor. The new challenge is the present paper is to accurately calculate the volume fraction of the gas generated within the molten slag and understand how the proposed geometry affects production, via the instantaneous thermodynamic equilibrium assumption. The main objective of this CFD study is to extract conclusions regarding the reactive slag flow, the extent of space utilization and the existence of dead volumes, and to provide design guidelines. A steady state sensitivity analysis of state variable distributions (namely, potential, temperature, velocity and gas volume fraction) with respect to a key design variable (the imposed voltage profile) reveals the reactor heating potential, the geometry of the Al region and the nontrivial operation, design and optimization problems.