Many sectors of the chemical industry suffer from a production overcapacity. Efficient production strategies are important. This situation combines with dropping computer costs to form an excellent environment for on-line optimization. If the transient periods of an operation are short relative to the steady intervals, a major part of the operation economics is determined by the steady state. Therefore, an optimization limited to the steady state will cover the main part of the attainable profit of the plant. A structure for an on-line optimizer is proposed. The optimization is conceived as a calculation of a set of optimal setpoints for the plant. The on-line optimizer is composed of a number of modules. The most important modules perform the optimization and identify the model. Although steady state models are much morre readily available and accurate than dynamic models, they still are approximate and contain parameters that have to be updated regularly to correct for the plant model mismatch and for slow changes in the plant. Sensitivity analysis of the optimization results and statistical analysis of the model identification results are combined in short-cut feasibility studies and on-line accuracy estimation. Data reconciliation improves the robustness of the application. Two examples serve as illustrations. The first example concerns a propane-propylene splitter. This example shows many of the interesting issues on a system of reduced size. The results are therefore easier to interpret. The second example is a boiler load allocation problem. This example is more involved and shows a realistic application.