The aquifer beneath an abandoned refinery in the Lower Rhine area, Germany, was contaminated with a number of different mineral oil products. Groundwater sampling in the area around the former xylene plant revealed that a xylene plume had developed in the underlying groundwater, and moreover, that there is strong evidence for in situ microbial xylene degradation with oxygen, nitrate, sulfate and ferric iron as electron acceptors. In order to prevent further xylene spreading, three pumping wells extracting contaminated water were installed downgradient of the spill zone. The numerical reactive transport code Transport Biochemisty Chemistry (TBC) was applied to this situation to quantify the relation of microbial degradation to xylene removal by the pumping wells. It could be shown that the unamended in situ degradation was an appreciable xylene removal process that contributed to about one-third to the total xylene removal (degradation plus extraction). A further objective of the model application was to predict xylene spreading under regional flow conditions, i.e. without operation of the three pumping wells, to consider the possible effects of natural xylene attenuation. To accomplish this, the model calibrated for the situation with operating wells was transferred to the hydraulic situation of regional flow while retaining the parameters of the biochemical model. It turned out that the xylene plume that is expected to develop downgradient of the source area will be limited to an extension of not more than 1000 m. An interesting feature of the simulations results was that xylene degradation under iron-reducing conditions, which was of minor importance for the situation with operating pumping wells, becomes the dominant degradation mechanism under regional flow conditions. Moreover, iron reduction will be the key process in controlling plume evolution. The model application illustrates that multi-species reactive transport models are needed to adequately transfer reactive processes from one hydraulic situation to another, while single species models are not suited for this predictive task.