Continuous powder mixing is gaining interest in the industrial community concerned with more and more functional powder products. The understanding of powder flow and mixing/segregation of particles as well as their translation into models that can be used in process monitoring and control is a major issue. In the present work, we describe the development of different mesoscopic Markov chain models that are based on interconnected compartments or cells delimited in the mixing chamber. The general structure of the chain allows the derivation of either homogeneous or non-homogeneous markovian models, for which transition probabilities are state-dependent. The models can be adapted to simulate variations of outflow rate, outlet mixture composition, hold-up weights and the distribution of these at the level of the compartments, during processing, including stationary and transitory phases. This is applied to a Gericke 500 GCM (R) continuous mixer for either pure powders or their mixtures, in the latter case through the consideration of a Markov chain for each component The models are fed by independent experiments that allow for the determination of the probabilities and the rules governing their change with the processing step, in particular during the transitory regimes. Agreement is found between model calculation and experimental data for a wide range of configurations. The models can catch the variations of hold-up weights and internal or outlet flow rates at any rotational stirrer's speed during mixer start and steady state. They can reproduce the variations of the outflow rate, and therefore mixture composition, when dealing with a mixture of two components. This is also presented for two nominal compositions. Conclusions are drawn in terms of process monitoring and control. It gives insights for process intensification, in particular for mixer design and the feeding configuration. (C) 2019 Elsevier B.V. All rights reserved.