PET with (18)F-FDG ((18)F-FDG PET) is increasingly used in the definition of target volumes for radiotherapy, especially in patients with non-small cell lung cancer (NSCLC). In this context, the delineation of tumor contours is crucial and is currently done by different methods. This investigation compared the gross tumor volumes (GTVs) resulting from 4 methods used for this purpose in a set of clinical cases. Data on the primary tumors of 25 patients with NSCLC were analyzed. They had (18)F-FDG PET during initial tumor staging. Thereafter, additional PET of the thorax in treatment position was done, followed by planning CT. CT and PET images were coregistered, and the data were then transferred to the treatment planning system (PS). Sets of 4 GTVs were generated for each case by 4 methods: visually (GTV(vis)), applying a threshold of 40% of the maximum standardized uptake value (SUV(max); GTV(40)), and using an isocontour of SUV = 2.5 around the tumor (GTV(2.5)). By phantom measurements we determined an algorithm, which rendered the best fit comparing PET with CT volumes using tumor and background intensities at the PS. Using this method as the fourth approach, GTV(bg) was defined. A subset of the tumors was clearly delimitable by CT. Here, a GTV(CT) was determined. We found substantial differences between the 4 methods of up to 41% of the GTV(vis). The differences correlated with SUV(max), tumor homogeneity, and lesion size. The volumes increased significantly from GTV(40) (mean 53.6 mL) < GTV(bg) (94.7 mL) < GTV(vis) (157.7 mL) and GTV(2.5) (164.6 mL). In inhomogeneous lesions, GTV(40) led to visually inadequate tumor coverage in 3 of 8 patients, whereas GTV(bg) led to intermediate, more satisfactory volumes. In contrast to all other GTVs, GTV(40) did not correlate with the GTV(CT). The different techniques of tumor contour definition by (18)F-FDG PET in radiotherapy planning lead to substantially different volumes, especially in patients with inhomogeneous tumors. Here, the GTV(40) does not appear to be suitable for target volume delineation. More complex methods, such as system-specific contrast-oriented algorithms for contour definition, should be further evaluated with special respect to patient data.