Di-methyl-ether (DME) is a solvent with clear potential for EOR through its novel use in DME enhanced water flooding (DEW). Fundamentally, this phase behaviour driven EOR process is based on an immiscible displacement of the oil phase by the water phase, enhanced by DME mass transfer between the two phases. Modelling such a process is far from trivial and requires a tailored modelling approach.In previous work (Chernetsky, et al., 2015) a dynamic model for DEW was successfully employed to history match core-flood experiments. This paper builds on this work by analyzing the driving mechanisms and sensitivities at different stages of the DEW process and describing the process of efficient upscaling of the model to field scale.The first part of this paper presents the basic modelling workflow for DEW for light to medium oils and typical concentration and saturation profiles and ternary diagrams. The objective is to determine the main driving mechanisms and their impact on the incremental oil recovery and DME utilization. The second part of the paper focusses on the upscaling of the DEW model to field scale. In particular, sensitivities to reservoir heterogeneity, relative permeability data and grid size are discussed through examples of sector and field models.The one-dimensional sensitivity studies provide valuable insight into the efficiency and dependencies of the fundamental recovery mechanism. The sensitivities focus on incremental oil recovery and DME efficiency and cover design parameters relevant to the DEW process, such as DME slug size and brine salinity, oil properties, remaining oil saturation and parameters used in simulation studies, such as grid size.Results for field scale modelling give guidance on how the DEW models can be effectively upscaled and what sensitivities can be expected at reservoir scale, and show the value of the model in supporting decision making and implementation of DEW in the field. Recommendations for field scale modelling are given, based on sensitivity analyses for various DEW scenarios. Computational techniques, such as adaptive gridding and parallel computing, are proposed to overcome limitations due to grid size sensitivity.