Functional magnetic resonance imaging relies on detecting small changes in the signal in the presence of noise from various sources. It has been shown that periodic variations in the signal at the respiratory frequency occur in the brain and various techniques have been proposed to remove them. However, the precise mechanism by which respiration affects the fMRI signal has not yet been proven. Here, we explore the nature of respiratory signal variations and the artefacts they produce in brain images. Our results demonstrate conclusively that bulk susceptibility variations in the lungs during respiration cause variations in the static magnetic field within the brain tissue. These variations in field strength and homogeneity lead to a shift of the image and a shading of image intensity in the phase encoding direction. These artefacts, if left uncorrected, may lead to the production of spurious activations and/or decreased statistical significance of true activations in fMRI. In addition, these results suggest that respiration effects may not necessarily be well characterized as simple additive noise and that an alternative model based on the physical origins of susceptibility variations may be more appropriate.