Process nonlinearity, multiple operating modes and time-varying characteristics often deteriorate the prediction performance of process models. In this article, a multi-mode moving-window Gaussian process regression (MWGPR) based approach for ARX modeling is proposed to effectively capture process nonlinearity or switching dynamics. The Gaussian mixture model (GMM) is first introduced to separate the data into different operating modes. Then the MWGPR strategy is applied to identify the local ARX model. Just-in-time learning (JITL) and dual updating are applied for more effective tracking of process dynamics. A simulation of a continuous fermentation process and a pilot scale experiment are presented to demonstrate the effectiveness of the proposed method. (C) 2016 Published by Elsevier Ltd.