Forecasts suggest that production of bitumen from oil sands reservoirs will increase by a factor of at least 2.5 times over roughly the next 15 years. Although a significant economic benefactor to the Canadian economy, there are challenges faced by oil sands operators with respect to greenhouse gas emissions and water consumption. For the Athabasca deposit, the current oil recovery process of choice is the SAGD (steam-assisted gravity drainage) method where high temperature and high pressure steam is injected into the oil sands formation. At present, there are more than ten SAGD operators in Alberta, Canada and results to date reveal that the geology of the reservoir impacts SAGD performance. Given the growth of the SAGD industry in Alberta, forecasting tools are required that can predict performance versus reservoir characteristics. Here, we present a neural network-based model to predict the SOR (steam-to-oil ratio) in oil sands reservoirs by using log and core data to characterize the reservoir porosity, permeability, oil saturation, depth and thickness. Our analysis confirms that the lower the porosity, permeability, and oil saturation of the reservoir, the worse the performance of the operation. In other words, the lower the quality of the reservoir, the lower the oil rate, and the higher the SOR. Our analysis also shows that well performance (i.e., SOR), is predictable with a relatively high degree of accuracy (R-2 similar to 0.80) using log and core data via a neural network model. These results imply that the depth of the reservoir, gamma ray readings, and permeability are the most important determinants of the variation in SOR. (C) 2015 Elsevier Ltd. All rights reserved.