Abstract Background In contemporary positron emission tomography (PET)/computed tomography (CT) scanners, PET attenuation correction is performed by means of a CT-based attenuation map. Respiratory motion can however induce offsets between the PET and CT data. Studies have demonstrated that these offsets can cause errors in quantitative PET measures. The purpose of this study is to quantify the effects of respiration-induced CT differences on the attenuation correction of pulmonary 18-fluordeoxyglucose (FDG) 3D PET/CT in a patient population and to investigate contributing factors. Methods For 32 lung cancer patients, 3D-CT, 4D-PET and 4D-CT data were acquired. The 4D FDG PET data were attenuation corrected (AC) using a free-breathing 3D-CT (3D-AC), the end-inspiration CT (EI-AC), the end-expiration CT (EE-AC) or phase-by-phase (P-AC). After reconstruction and AC, the 4D-PET data were averaged. In the 4D_avg data, we measured maximum tumour standardised uptake value (SUV)_max in the tumour, SUV_mean in a lung volume of interest (VOI) and average SUV (SUV_mean) in a muscle VOI. On the 4D-CT, we measured the lung volume differences and CT number changes between inhale and exhale in the lung VOI. Results Compared to P-AC, we found −2.3% (range −9.7% to 1.2%) lower tumour SUV_max in EI-AC and 2.0% (range −0.9% to 9.5%) higher SUV_max in EE-AC. No differences in the muscle SUV were found. The use of 3D-AC led to respiration-induced SUV_max differences up to 20% compared to the use of P-AC. SUV_mean differences in the lung VOI between EI-AC and EE-AC correlated to average CT differences in this region (ρ = 0.83). SUVmax differences in the tumour correlated to the volume changes of the lungs (ρ = −0.55) and the motion amplitude of the tumour (ρ = 0.53), both as measured on the 4D-CT. Conclusions Respiration-induced CT variations in clinical data can in extreme cases lead to SUV effects larger than 10% on PET attenuation correction. These differences were case specific and correlated to differences in CT number in the lungs.