Respiration causes variations in the signals acquired during magnetic resonance imaging (MRI) and therefore is a significant source of noise in functional brain imaging. A primary component of respiratory noise may arise from variations of bulk susceptibility or air volume in the chest. Here we investigate the nature of the image artefacts that can be caused by such changes. We develop a simple model which attempts to mimic the effects of variations in susceptibility and volume during respiration. Theoretical calculations, computer simulations and imaging experiments with this model show that small variations in susceptibility within the thorax from alterations in the paramagnetism of cavity gas may lead to a shift of the image on the order of 0.1 pixels as well as a shading of the intensity by +/-1%. These effects are observed to be predominant in the phase-encoding direction. They may lead to the production of spurious activations in functional MRI and are likely to be of more importance at higher field strengths.