The purpose of this study was to parameterize TOF image quality improvement over non-TOF PET using small lesions in a tapering phantom that represents a range of body sizes. A set of 46 1.0-cm spherical inserts was divided into 6 groups and positioned in a fillable, tapered phantom that represents a range of body dimensions (an oval cross section from 38.5 cm x 49.5 cm to 6.8 cm x 17.8 cm with a length of 51.1 cm). The hot spheres were positioned in a uniform warm background (8:1 radioactivity concentration) and were distributed in 6 regions of different cross-sectional areas: 6 spheres at 306 cm2, and 8 spheres at 426, 574, 754, 964, and 1187 cm2. The PET/CT study was acquired on a Discovery 690 PET/CT (GE Healthcare), which uses LYSO crystals (4.25 × 6.3 × 25 mm3) in three rings of 6 x 9 blocks. Six frames, with a different set of spheres centered for each frame, were each acquired. Images were reconstructed with and without TOF information using 3D OSEM with 8 subsets and 1-20 iterations. The 50-cm field of view (FOV) was reconstructed to a 128x128 matrix with attenuation, normalization, decay, scatter, singles-based random corrections, and z-axis smoothing, but no post-filter was applied. An automated algorithm was used to locate and place regions of interest on the hot spheres in the CT, and then apply the ROIs to the PET images to measure signal and background variability (image noise). When both TOF and non-TOF images are near convergence, the ratio of contrast recovery for TOF to non-TOF is ∼1.0 for small body sizes, but the CRC increase ranges from 10%-25% for body sizes with diameters >30.0 cm. When comparing the small lesion signal-to-noise ratios between TOF and non-TOF images, the square of the SNR ratios for TOF to non-TOF was a function of body size and timing resolution with an offset of -1.36. ©2009 IEEE.