This paper describes the application of machine-learning techniques for unlocking the full potential of nuclear magnetic resonance (NMR), using a case study from the Seven Heads gas field. This field has long been recognized but was not developed due to a variety of technical challenges, including the thin-bedded nature of the sediments and the presence of both mobile and immobile viscous residual oil. The oil is a highly viscous liquid which, if produced, could block production tubing due to the shallow depth of the reservoir and associated low pressures. To produce dry gas successfully, identification of both oil and gas zones was necessary to enable gas zones to be perforated and oil zones to be excluded. During the development drilling campaign, the reservoir was appraised using a formation evaluation program specifically designed to address the presence of oil within the thinly bedded reservoir. In conjunction with core data and high-resolution electric logs, NMR logs were used to identify and avoid perforating zones with higher oil saturations. Formation fluid types were derived from the NMR using a pattern recognition technique that analyzes the entire shape of the T1 and T2 distributions to derive the volumes of gas, oil, and water. This machine-learning technique was calibrated using Dean and Stark fluid analysis data and enabled the prediction of continuous water, gas, and oil saturation curves. The results were used to ensure that the perforation strategy avoided oil-bearing sands. This paper describes how the NMR, together with machine learning, has enabled a complex tight gas field to be developed.