In the recent decade, automatic history-matching has been a very active research area with significant progress reported. Workflows have been designed in order to build reservoir models consistent with production history and geology. Due to the growing use of repeated seismic and monitoring, it has become important to incorporate also 4D seismic attributes in the modeling of reservoirs. More complicated workflows have been proposed to simultaneously integrate production history and 4D-seismic data in reservoir models. They encompass the minimization of an objective function, which measures the mismatch between the collected data and the corresponding simulated answers. Thus, the objective function is used to compare the actual production data to the simulated ones and the actual 4D seismic attributes to the simulated ones. The computation of 4D seismic answers entails the use of the simulated fluid distributions. The flow simulations, which provide these fluid distributions, are usually performed at the coarse scale. A downscaling process is required to properly model the 4D seismic answers. In this paper, we propose a new method for downscaling pressures and saturations from the coarse reservoir scale to the fine geological scale. This method ensures mass conservation and allows capturing fine scale heterogeneity. It calls for local flow simulations with boundary conditions specified from the driving coarse flow simulation. A numerical example is presented to point out the improvement in the building of the constrained reservoir model.