The present invention is a method for determining the optimum operating practice for an industrial process. The methodology includes a combination of process control steps and data analysis techniques to investigate responses to controlled "black box" process inputs, e.g., coal and air feeds for coal-fired furnaces in the case studies above. A unique time-dependent perturbation is applied to individual process feed streams to give a signature in the data that permits the filtering and detection of various measured responses, e.g., continuous emission monitor readings for NOx, SO2, CO2, CO, and opacity for regulated electric utility furnaces. The perturbation is implemented using the plant control system. Individual control signal set points are perturbed for each piece of equipment supplying process inputs of interest. As a result, complex commercial processes can be evaluated for optimization without changing the normal configuration of the system.