A performance modeling of phosphoric acid fuel cell was constructed using error back-propagetion neural network with generalized delta rule. The network was trained to produce the performance curve according to cell temperatures, H₂ flow rates and O₂ flow rates. The reliability of performance prediction was verified by comparing the experimental data of unit cell. The three layered error back-propagation net-work learned exactly the performance curve of unit cell when the step size coefficient was 0.7 and the momentum was 0.9. It offered reasonable prediction for various O₂ and H₂ flow rates and temperatures. This neural network combined to commercial process simulator ASPEN PLUS as a unit for fuel cell stack. In this process simulation, 50kW PAFC system was selected, and the effects of operating variables on the performance of the system were also investigated. The maximum power of fuel cell was achieved when the reformer temperature was 600℃ at 1 atm, and steam-carbon ratio was 2.6. It is proper to maintain the operating temperature of fuel cell at 190℃.