Abstract: A prediction model of depropanizing column feed composition was established.The prediction model was decomposed into an optimization model and a process simulation model, which made it easy to be solved.The process simulation model was solved by using the process modeling simulation software, and the optimization model, a constrained NLP (non linear programming) problem, was simplified so that it could be solved by simplex search and genetic algorithms.The communication between optimization model and process model was carried out through the interface program of process modeling simulation software and the optimization method.A case study on the depropanizing column of a refinery showed that the proposed model and approach could accurately and quickly predict depropanizing column feed composition, and have a good application potential in pretrolenm refining and petrochemical industry.