Eight different modifications of the same single tissue compartment model to measure myocardial blood flow, based on inhalation of 15O-labelled CO2 and positron emission tomography, were assessed in both dogs and human normal volunteers. Several models provided results with the same degree of accuracy in dogs. However, a number of these models gave poorer results in humans. It was established that the model containing components for blood flow, fraction of water exchanging tissue and spill-over arterial blood volume provided the most accurate and reproducible results. This model contains inherent corrections for the limited spatial resolution of positron emission tomographs. For ease of computation, linearisation of the operational (fitting) equation was tested, but found not to be satisfactory. The left atrium was slightly better than the left ventricle for determining the arterial input function. Inclusion of the blood volume term in the fitting procedure was significantly better than subtracting blood volume prior to analysis, both in terms of accuracy and precision.