A strategy of developing on-line optimization intelligent systems based on combining flo wsheeting simulation and optimization package with artificial neural networks (ANN) is presented in this paper. A number of optimization cases for a certain chemical plant are obtained off-line by using PROCESS-Ⅱ or other flowsheeting programming with optimization. Then, taking these cases as training examples, we establish a neural network systems which can be used on-line as an optimizer to obtain setpoints from input data sampled from distributed control system through gross errordetecton and datareconciliation procedures./Such an on-line optimizer possesses two advantages over nonlinear programming package: first of all, there is no convergence problem for the trained ANN to be used online; secondly, the frequency for setpoints updating is not limited because only algebraic calculation rather than optimization is required to be carried out on-line. Here two key problems of implementing ANN approaches to the on-line optimization are discussed: how to improve the prediction accuracy of ANNs models for meeting the optimization requirements. Results from an actual fractionation unit of a FCC plant in a refinery showed a 0.5%-1.0% increase in the total recovery of light oil products. Details of the strategy used are described.