Low-field nuclear magnetic resonance (NMR), whether implemented in a logging tool, bench top analyser or on-line sensor, cannot detect the complete response of heavy oil or bitumen. Both heavy oil and bitumen relax quickly so the majority of these oils' spectra are detected at relaxation times less than 10 ms at room temperature. In clay-free sands, the contribution of heavy oil to the NMR spectrum is distinct and, as a result, it is still possible to calculate oil and water content based on NMR spectra. However, in sands that contain clays, the relaxation times of clay bound water are in the same range as bitumen. Experimental results from mixtures containing illite, kaolinite, montmorillonite, sand and mild brine show that clay bound water has a characteristic response. These NMR 'signatures' were used to develop predictive nomographs of clay content. A second set of experiments involved adding heavy oil to mixtures containing clay, sand and brine. The changes in NMR spectra after exposure to heavy oil were compared to the spectra obtained before oil was added. The differences identified in this work allowed for improvements in calculating water, oil and/or solids content. This paper presents a preliminary predictive algorithm for clay content determination, and this knowledge will allow one to more accurately separate the contributions of heavy oil and clay bound water from a sample despite the fact that these will overlap in an NMR spectrum. Improved characterization of oil sands is a possible consequence of this work.