This paper proposes a new response surface building method for accurately estimating reservoir flow performance. Usually, due to time consuming calculation, the allowed number of flow simulations does not provide enough information for capturing the curvature and non-linearities in the flow response surface. The method combines classical flow simulation results with secondary information provided by fast flow simulation such as the time-of-flight connectivity. For this purpose, the method is based on the discrete smooth interpolation algorithm which has the advantages to be (1) data exact, (2) unbiased by the choice of an initial model, (3) flexible, so complex response surface can be modeled, and (4) can handle in a unified and simple framework heterogeneous data. A modified version of the SPE 10th Benchmark Model has been used to investigate the accuracy of the generated response surface, and monte-carlo sampling was carried out to measure the impact of the method on uncertainty assessment. Results show that the proposed method (1) works accurately and efficiently in modeling the repones non-linearities and (2) provides a satisfactory P10, P50 and P90 values for uncertainty assessment.