Because anatomical information on fluorine-18 fluorodeoxyglucose (FDG) whole-body positron emission tomography (PET) images is limited, combination with structural imaging is often important. In principle, software co-registration of PET and computed tomography (CT) data or dual-modality imaging using a combined PET-CT camera has an important role to play, since "hardware-co-registered" images are thereby made available. A major unanswered question is under which breathing protocol the respiration level in the CT images of a patient will best match the PET images, which represent summed images over many breathing cycles. To address this issue, 28 tumour patients undergoing routine FDG PET examinations were included in this study. In ten patients, PET and CT were performed using a new combined high-performance in-line PET-CT camera without the need for repositioning of the patient, while in 18 patients imaging was performed on separate scanners located close to each other. CT was performed at four respiration levels: free breathing (FB), maximal inspiration (MaxInsp), maximal expiration (MaxExp) and normal expiration (NormExp). The following distances were measured: (a) between a reference point taken to be the anterior superior edge of intervertebral disc space T10-11 and the apex of the lung, (b) from the apex of the lung to the top of the diaphragm, (c) from the apex of the lung to the costo-diaphragmatic recess and (d) from the reference point to the lateral thoracic wall. Differences between CT and corresponding PET images in respect of these distances were compared. In addition, for each of 15 lung tumours in 12 patients, changes in tumour position between PET and CT using the same protocol were measured. CT during NormExp showed the best fit with PET, followed by CT during FB. The mean differences in movement of the diaphragmatic dome on CT during NormExp, FB, MaxInsp and MaxExp, as compared with its level on PET scan, were, respectively, 0.4 mm (SD 11.7), -11.6 mm (13.3), -44.4 mm (25.5) and -9.5 mm (25.6). CT acquired in MaxExp and MaxInsp is not suitable for image co-registration owing to the poor match of images in MaxInsp and because of difficulties with patient performance in MaxExp. With reference to lung lesions, NormExp showed the best results, with a higher probability of a good match and a smaller range of measured values in comparison with FB. Image misregistration in combined PET-CT imaging can be minimized to dimensions comparable to the spatial resolution of modern PET scanners. For PET-CT image co-registration, the use of a normal expiration breath-hold protocol for CT acquisition is recommended, independent of whether combined PET-CT systems or stand-alone systems are used.