A new optimal control technique is presented to provide good quality, robust solutions for chemical engineering problems, which are generally non-linear, and highly constrained. The technique neither uses any input of feasible control solution, nor any auxiliary condition. The technique generates optimal control by applying the genetic operations of selection, crossover, and mutation on an initial population of random, binary-coded deviation vectors. Each element of a deviation vector corresponds to a control stage, and is a deviation from some "mean" control value randomized initially for that stage. The deviation, and the mean control value map on to the actual discrete step value of control at that stage. The mapping is logarithmic in beginning, but is later allowed to alternate with a linear one. The genetic operations are periodically followed by the replacement of mean control values by a newly available optimal control solution, and by the size-variation of control domain between its limits. The optimal control technique is successfully tested on four challenging problems of chemical engineering. (C) 2003 Elsevier Ltd. All rights reserved.