Hydrocarbon production prediction, as it is important for the evaluation of a petroleum company, it is as complex as well, considering it depends on numerous input parameters. Increase in the model compexity increases the number of parameters influencing the simulation results significantly. Uncertainties in different data, whether seismic, well logging, laboratory or well testing multiply the uncertainties of simulation models so it is not advisable to speak of exactness but only about certain level of accuracy or precision. Number of simulation gridblocks is decided upon available data among others, and while smaller number of gridblocks means shorter numerical simulation, it can mean lesser accuracy as well. However, if the model is extremely refined, and the data available is rather coarse, simulation errors increase. That is why it makes sense to find a relation between the costs of extra laboratory and field testing and the redcution of the difference between the simulation analysis and real data. Two hypotheses arise: (1) it is possible to quantify the difference between numerical simulation and real data by a statistical analysis of the input and output parameters and (2) additional investment in more detailed reservoir and fluid testing can be statistically justified by a proof of an increased precision of production prediction obtained by simulation. Due to already mentioned uncertainties, it is usual to resort to the more intuitive, deterministic approach of estimating reserves, which can lead to great financial losses as the whole economics is calculated based on only one figure for proven reserves and the risks of various but possible scenarios are not accounted for. It is therefore advisable to take the parameters probability distributions of the reservoir in question into account but more important is to address the range of those parameters values. After the available data have been statisticaly analysed, it was concluded that they were valuable guidelines for the distributions ranges used in the numerical simulations. The influence of model dimensions (model resolution), porosity, irreducibile water saturation, fluid contact depth, capillary pressure and relative permerability curves, fluid momdel, permeability and reservoir heterogeneity on history matching have been examined first, and then their influence on future production was tested. The economic aspect of the precision increase was analysed through measurements relative prices of each influential parameter. Computer neural networks were used for parameter influence determination, and for the economics analysis it was assumed that the well logging in a new well is the most expensive item. Grid influence was tested through 9 different resolutions considering other things equal, and it was obvious that the lowest resolution grid could not achieve the real cumulative production, which can be mostly ascribed to an unrealistically low vertical resolution (number of gridblocks in the z-direction). The differences in various grids become more obvious when considering each well bottom hole pressures, although a significant increase in grid resolution did not lead to significantly smaller error in simulation versus real data. The porosity influence was tested on 5 models, and the difference in the cumulative production could be only seen in the lowest resolution grid that could not achieve the reported cumulative production even when the porosity was increased by 67%. The differences are more expressed if the bottom hole pressures are observed, and one of the wells even has a mismatch in gas rate in the last period od production. The fluid model influence is more pronounced in lower resolution grids, especially in the grid with an unrealistically high ratio of vertical and horizontal resolution. Setting the fluid contact 2 m deeper compared to the official depth impacts the bottom hole pressures in lower resolution grids compared to higher resolution ones, just as setting the contact 2 m above the reported depth. Selection of other permeability distributions showed to be very influential – considering bottom hole pressures in all wells, it can be concluded that the differences between the distributions are smaller in the lower resolution grid. Vertical permeability variation influence is less pronounced. Inflence of rock compressibility was tested and there was not much deviation considering the cumulative production of the reservoir. The difference between the deviations of two higher resolution grids, among which one can be considered relatively coarse, is smaller compared to the difference between the deviation in two smaller resolution grids. Finally, except for some extreme cases, relative permeability and capillary pressure curves combinations, which are dependant on irreducibile water saturation, showed less influence on cumulative production and bottom hole pressures, but they define the share of water and condensate in the overall liquid production. Running a neural network showed that the impact significance of each parameter differs from well to well, meaning that it is crucial to characterize some parts of the reservoir in more detail as some wells showed to be more sensitive to a grid change, while others are highly sensitive to reservoir characteristics as horizontal permeability and rock compressibility. This was also confirmed in the prediction cases, where porosity showed to be a parameter important for bottom hole pressure, while grid refinement is still the most important parameter for cumulative production.