Chromatographic processes provide a powerful tool for the separation of mixtures, in which the components have different absorption affinities. In this area, the simulated moving bed (SMB) technology is becoming an important technique for large-scale continuous chromatographic separation processes and its applications are found in the pharmaceutical industry and in the production of fine chemicals. The successful design and operation of SMB units depend on the correct choice of operating conditions, particularly of the flow rate in each zone. A detailed and reliable dynamic model of the process is necessary for proper design; however, due to the complex dynamics of the process, the choice of operating condition is not straightforward. In this work, the iterative learning design based on partial least squares (PLS) approach for SMB process is proposed. PLS is a data-driven approach that extracts the essential information of the process state from operation data through mathematical and statistical methods. It facilitates data-compression by condensing the variance of the process into a very low-dimensional latent subspace. Moreover, the MIMO SMB process can be transformed into a system of the parallel uni-variate SISO design problem. Also, the incorporated dEWMA strategy acts as a model updating method to reflect process decay or drift. Based on the dEWMA PLS model, two design strategies are proposed: cycle-to-cycle and within cycle design. The performance of these design strategies is assessed by implementation on a virtual eight column SMB unit. (C) 2008 Published by Elsevier B.V.