A compositional simulator that is fully compositional and more robust and accurate is disclosed. Relative permeability (kr) and capillary pressure (Pc) are modeled as state functions, making them unique for a given set of inputs, which can include Euler characteristic, wettability, pore connectivity, saturation, and capillary number. All of these are made to be a function of composition, T, and P or rock properties. These state function kr-Pc models are fully compositional and can fit experimental data, including complex processes such as hysteresis. The models can be tuned to measured relative permeability data, and then give consistent predictions away from that measured data set. Phase labeling problems are eliminated. Flux calculations from one grid block to another are based on phase compositions. Simulations for three or four-phase hydrocarbon phases are possible. Time-step sizes increase to stability limits of implicit-explicit methods. Fully implicit methods are possible and significant improvements are expected.