The aim of my thesis is to examine the variances in the flow properties of stochastic realisations of fracture patterns, also with regard to model size. The importance of studies on naturally fractured reservoirs has the background that a third of the worldwide oil and gas reserves are located in such reservoirs. The accurate assessment of NFR is the key to develop a reservoir model. One problem is that there is no REV in fractured media, which is the result of the high heterogeneity of NFRs. To get a lower uncertainty it is necessary to not just include the fractures in the flow modelling process but also the rock matrix, which examines the importance of DFM modelling. To compute the permeability tensor flow based up scaling was used. My thesis focuses on the determination of equivalent permeability and their anisotropy. To analyse the uncertainty, which is induced in the stochastic generation process of the models, I generated 50 equi-probable realisations of fracture sets, followed by an analysis of randomly chosen combinations of multiple fracture sets and a cluster analysis. The model size required depends on fracture length statistics, as the model has to be bigger than the mode of the fracture size to avoid a high uncertainty. To find the smallest permissible model size, to still have an acceptable computation time in the meshing and simulation process and also to have the lowest possible error, it is necessary to make a short statistical analysis such as in my provided workflow. I also compare the flow simulation results using the commercial software FracMan by Golder Associates with extended fracture generation algorithms, implemented in a Rhinoceros 5.0 plug-in by L. Mosser (2013) called StatFrac, that honour the “forbidden-zones” around fractures. A check during the fracture set generation process is made, to assure that there are no other fractures generated in the zone of relieved stress of the fracture. Following my statistical analysis of the fracture sets and their combinations, I calculated equivalent permeability using parallel-plate law. I used a constant matrix permeability for every model and a fracture permeability which is constant within each fracture but varies for every single fracture using a configuration file. The effects of the uncertainty, which appear while randomly generating the models using the different stochastic fracture generation algorithms, are also visible in the range of my simulation results. Furthermore the impact of the orientation of the fractures on the flow has been studied. Expectedly, fracture sets aligned with the far field fluid pressure gradient have a higher velocity than perpendicular ones. To investigate the influence of the matrix permeability on equivalent permeability and flow velocity, I carried out a sensitivity analysis. The outcome is, that the higher the matrix permeability is, the higher equivalent permeability gets. The flow velocity in the fractures increases, due to the interaction between matrix and fracture sets. The matrix contributes to the flow starting with the lowest permeability used (5 mD). Rock matrix would dominate the flow with a permeability higher than 500 mD.