Purpose: To examine the efficacy and accuracy of respiratory‐gated positron emission tomography (4D‐PET) maximum intensity projections (MIPs) in segmenting mobile tumor internal target volumes (ITVs) through a phantom study. Methods: An acrylic phantom was used for PET list‐mode acquisition, consisting of 11C‐filled spheres affixed inside a cylinder containing an 18‐fluorodeoxyglucose solution. The phantom was attached to a robotic arm that underwent 1D motion based on clinically‐derived patient breathing trajectories. An amplitude‐based gating was performed on sequential list‐mode sub‐files of varying signal‐to‐background ratios, and PET‐MIPs were generated from the gated images. ITVs were segmented by first denoising and deblurring images using a custom post‐processing module and then applying an absolute SUV threshold. ITV accuracy was assessed using the volume recovery fraction (VRF) ‐ the ratio of measured ITV to true volume occupied by the moving phantom spheres ‐ which was used to compare PET‐MIP, ungated PET, and static sphere images. In addition, the effects of tumor trajectories, number of gating windows, and margin additions were investigated. Results: VRFs of ITVs generated from PET‐MIPs were consistently higher than those from ungated PET. They also demonstrated a closer agreement to VRFs of static spheres (up to 99% similarity vs. 72% for ungated PET), suggesting that tumor motion had very little effect on the accuracy of PET‐MIP measurements. Trajectories with higher amplitude and baseline drift decreased VRF by up to 14% for both PET‐MIP and ungated PET. Increasing the number of gating windows (up to eight windows) resulted in higher VRFs for PET‐MIPs without producing excessive image noise. PET‐MIPs required a smaller margin addition than ungated PET to achieve a better overlap with the ground truth ITV. Conclusions: Compared to ungated PET, PET‐MIPs are not significantly affected by tumor motion. 4D‐PET imaging is a promising and viable methodology to better delineate mobile tumor volumes.