Distributed Model Predictive Control (DMPC) algorithms reported in the literature had often assumed that state measurements are available to all subsystems at all time instants. However, this assumption is not valid in the presence of loss of communication or data. Then designing for such losses within DMPC becomes critical to overall performance and stability of the system. In this work communication loss was considered for robust DMPC by using within each subsystem an observer, which continuously and recursively estimates states' bounds for all the plant states. These state bounds are then posed as extra constraints within a Linear Matrix Inequalities' formulation that is used to calculate the control actions based on the minimization of an upper bound on robust performance. (C) 2016 Elsevier Ltd. All rights reserved.