Purpose: Sphericity has been proposed to characterize PET tumor volumes, with complementary prognostic value with respect to SUV and volume in both head and neck and lung cancer. The objective of the present study was to investigate its dependency on the tumor delineation and the resulting impact on its prognostic value. Materials and methods: Five segmentation methods were considered: 2 thresholds (40% and 50% of SUVmax), the ant colony optimization (ACO), the fuzzy locally adaptive Bayesian (FLAB), and the gradient-aided region-based active contour (GARAC). The accuracy of each method to extract sphericity was evaluated on a dataset of 176 simulated, phantom and clinical PET images of tumors with associated ground-truth. The prognostic value of sphericity and its complementary value with volume for each segmentation method was evaluated in a cohort of 87 stage II-III lung cancer patients. Results: Volume and sphericity associated values were dependent on the segmentation method. The correlation between the segmentation accuracy and the sphericity error was moderate (|ρ| from 0.24 to 0.57). The accuracy in measuring the sphericity was not dependent on the volume (|ρ|<0.4). In the lung cancer cohort, sphericity had prognostic value although lower than volume, except for FLAB for which a small improvement over volume alone when combined with sphericity was observed (hazard ratio of 2.67 compared to 2.5). Substantial differences in patients’ prognosis stratification were observed depending on the segmentation. Conclusion: The tumor functional sphericity was found to be dependent on the segmentation method, although the accuracy to retrieve the true sphericity was not dependent on tumor volume. In addition, even an accurate segmentation can lead to an inaccurate sphericity, and vice-versa. Sphericity had similar or lower prognostic value than volume in the NSCLC cohort, except with one method (FLAB) for which there was a small improvement in stratification when combining both parameters.