Solids circulation rate is an important parameter that is essential to the control and improved performance of a circulating fluidized bed system. The present work focuses on the identification of a cold flow circulating fluidized bed using a multiple model identification technique that considers the given set-up as a nonlinear dynamic system and predicts the solids circulation rate as a function of riser aeration, move air flow rate, and total riser pressure drop. The predictor model obtained from this technique is trained on glass beads data sets in which riser aeration and move air flow are varied randomly one at a time. The global linear state space model obtained from the N4SID algorithm is trained on the same data set and the prediction results of solids circulation rate from both these algorithms are tested against data obtained at operating conditions different from the training data. The comparison between the two methods shows that the prediction results obtained from the multiple model technique are better than those obtained from the global linear model. The number of local models is increased from two to five and two third order state space models are sufficient for the present sets of data.