Knowledge of oil viscosity is important when estimating hydrocarbon reserves, and evaluating the potential for water-flooding or EOR processes. This information is especially important in heavy oil and bitumen, as viscosity is usually the major impediment to recovery of these reserves. As oil viscosity increases, obtaining a laboratory measurement is difficult and prone to error, and viscosities measured in the lab may not be representative of field conditions. Nuclear magnetic resonance (NMR) is therefore presented as an attractive alternative method for determining oil viscosity. Several correlations already exist for determining oil viscosity using NMR. Some of these correlations compare the geometric mean T-2 relaxation time to oil viscosity, while others relate viscosity to the apparent hydrogen index. This paper examines these different models on a suite of conventional and heavy oil samples. It is concluded that none of the existing models can accurately predict oil viscosity for both conventional and heavy oils, especially for oils with viscosity higher than 20,000 cP. All the measured oil samples show a correlation between oil viscosity and the geometric mean T-2 relaxation time, and also between viscosity and relative hydrogen index. This is consistent with what other experimenters have noticed. An empirical model is developed, correlating oil viscosity to both of these parameters. Unlike previous models, this model can accurately predict oil viscosity for both conventional and heavy oil. The wider range of this model makes it useful for laboratory analysis of oil viscosity using NMR. If the results of this model can be applied to in situ oils, NMR can be used as a logging tool to characterize heavy oil and bitumen formations. The model presented in this paper is the first step towards successfully predicting viscosity in situ.