In the oil and gas industry, numerical reservoir modelling and simulation is the most common and widely used tool for decision making. However, construction of such model requires a large amount of data from sources such as well logs, seismic, core analysis and well testing. The data obtained are not error free, they are prone to measurement error. Moreover, due to the humongous size of reservoirs, their properties are inferred from measurements taken at only some limited locations within the vicinity. Therefore, calibration of numerical reservoir models is necessary. In this paper, artificial neural network modelling is applied for calibration of a numerical reservoir model under primary recovery. A benchmark reservoir model, referred to as PUNQ-S3, was selected to investigate the application of neural network model. The neural network model was developed by using an inverse solution method to formulate the training and testing data. From the numerical reservoir simulation point of view reservoir properties to be calibrated are the inputs while pressure and production profiles are outputs. On the contrary, we have employed the inverse. This allowed carrying out history matching without the requirement of an objective function and optimization algorithm. In order to calibrate the numerical reservoir model, the trained neural network was simulated by using historical data. The results of this study have brought forward an improved technique for history matching (calibration) of a numerical reservoir model. © 2018 � IOS Press and the authors. All rights reserved.