Purpose: This work systematically evaluates the effect of source‐to‐background ratio, tumor size, scan duration, and smoothing filter on a novel two‐stage PET tumor segmentation method in phantoms. The two‐stage method uses an adaptive region‐growing algorithm and a dual‐front active contour model, resulting in advantages of reproducibility and no need of choosing a threshold. Method and Materials: Six spherical tumor phantoms (0.5 – 20 mL) in a warm cylindrical container were scanned for 120 minutes in a PET/CT scanner. The source‐to‐background (S/B) ratio ranges from 16 to 0.5. For every 2‐min scan, three PET images were reconstructed using the iterative Ordered Subset Expectation Maximization (OSEM) algorithm with 5mm‐FWHM, 2mm‐FWHM post‐reconstruction smoothing filters and without the filter, respectively. At S/B ratio 2, images for longer scan durations were also generated. The two‐stage method was applied to segment the tumor phantoms with optimized parameters. Results: The segmentation accuracy depends mainly on the S/B ratio and tumor size, with a higher accuracy for larger tumor or higher S/B ratio. The 5mm‐FWHM filter reduces more noise than the 2mm‐FWHM filter and the unsmoothed. This leads to better segmentation for images smoothed with the 5mm‐FWHM filter. Lengthening scan duration did not improve the segmentation for tumors ⩾ 6 mL, but made the imperceptible 1 mL tumor detectable. The overall differences were not statistically significant between scan durations or between OSEM‐2mm and unsmoothed, but were statistically significant between OSEM‐5mm and unsmoothed. Conclusions: We concluded that a computerized PET tumor segmentation method should incorporate these factors (S/B ratio, tumor size, scan duration, and smoothing filter) if known or estimable. Secondly, it is very important to use the same PET/CT protocol for consistent quantitative evaluation. Finally, longer scan duration improves small tumor detection in PET.