This thesis is investigating the feasibility and impact of delumping production well streams which are used for prediction of process plant products, for a gas condensate and a volatile oil as reservoir fluids, by using coupled reservoir-surface simulation. Because of the amount of equations which are solved during reservoir simulation, run-time is a crucial limitation. Therefore, it is commonly practised to lump components of a fluid model. In order to acquire detailed information about the system, delumping is applied afterwards. The foundation of this thesis are phase behaviour models for both reservoir fluids with detailed component information. These models were used to create reference solutions. The detailed fluid models were then analysed by fuzzy clustering in order to realize lumping schemes which were implemented using PVTi® as fluid modelling software. These lumped fluid models were simulated with a reservoir model computed by ECLIPSE® E300, using its delumping facility to create time dependent well streams with detailed compositions. These well streams serve as input for the process simulation model calculated by Aspen HYSYS®. The obtained process plant products such as sales gas, liquefied petroleum gas (LPG) and condensate, were compared with the reference results and deviations from such were analysed and interpreted. It was concluded that the impact of delumping compositional well streams for processing them in a gas treatment unit, to obtain process plant products, is almost negligible. The importance of a consistent lumping, and therefore sensitivity of fluid descriptions within a simulation environment, was proven. In addition, several, sometimes not well documented, limitations, of the available inbuilt delumping facility of ECLIPSE® E300, were found for using it in a coupled reservoir-process simulation approach. The procedure on how to create a coupled subsurface-surface simulation for a gas condensate and a volatile oil are reviewed in this work as well as recommendations and remarks on the tasks and its influences are proposed.