Development of low permeability reservoir is very different from that of normal reservoir. Complicated geological and engineering factors conspire to affect wells production performance. Moreover, these factors vary with the development strategies for different low permeability reservoirs. It is difficult to accurately describe these factors and further predict its production performance. This paper presents a new methodology integrating soft computing and hard computing to predict production performance of low permeability reservoir. Integration of data and knowledge is used for soft computing. Firstly, the knowledge about fluid flow is extracted by numerical simulation (a conventional hard computing technology). Then the knowledge together with corresponding data is input to model production performance by neural network (a popular soft computing technology). Lastly, the trained models are used to predict the production performance. This method is proved to be effective for low permeability reservoir.