Abstarct: Reservoir’s properties models that used to fluid flow simulation in porous media include millions of cells, which are not reasonably computational to solve such models with this cell volume. Upscaling methods are used to reduce the number of cells in the original model. The main idea of upscaling is to replace a coarse-scale homogeneous cell rather than multiple heterogeneous fine-scale cells. Several methods have been developed to scale up, which are baxsed on averaging. In this thesis, a robust method called a variable-width window, the main idea of is the bandwidth, is proposed as a scale-up method. The aim of this research is to simultaneous upscaling of the two features of the reservoir in two dimensions. Given the fact that in order to determine the number of cells in the simulator model, an appropriate division must be made according to the variability model, this approach can be used to determine the appropriate bandwidth. Since the distribution of the reservoir is different in various regions, the coarse-graining process should be done baxsed on the feature variability. By this method, regions with severe variability will remain fine and areas with variable variability will increase scales. The upscaling process has been investigated in four modes: upscaling of data in one dimension with one and two features of the reservoir, and then generalizing the process to two dimensions with one and two reservoir features simultaneously. The main challenge in the variable-width window procedure is determining the bandwidth or width of the window. In this method, the width of the window is due to the variable variability of the reservoir property, which is controlled by the bandwidth. In one-dimensional state, the bandwidth is scalar and in two dimensions, it is defined as a vector. To simultaneous upscaling of hydrocarbon reservoir features, two maximum and minimum bandwidth approaches have been used. The STUF models are completely identical in terms of structure, and their only difference is in the quantity of cells. In addition to the proposed method, the process of upscaling is also carried out baxsed on the wavelet transform. The result of upscaling of the two methods suggests that the simulator models obtained from the proposed method of thesis are more similar to the original model compared to the wavelet transform. This is evident in the comparison of the upscaling error of the upscaled models and the variance of the data. Upscaling baxsed on the variable-width window is a completely intelligent process and only baxsed on the cell's variability, while wavelet-baxsed upscaling is on the binary system. In a variable-bandwidth window approach, it will be possible to scale up the process by specifying the expected error and the cell numbers of simulator model.