The capabilities and limitations of two-(2D) and three-dimensional (3D) fluorine-18 fluorodeoxyglucose positron emission tomography (FDG-PET) in detecting small tumors and lymph nodes were studied in a phantom modeling the human chest and axilla. Multiple dual-radionuclide phantom studies were performed. Five hollow spheres ranging in diameter from 3 mm to 15 mm were filled with carbon-11 and placed in the axillary and mediastinal regions of an anthropomorphic phantom containing hollow organs filled with 18F to simulate FDG uptake 1 h after injection. Dynamic imaging was performed to acquire PET images with varying target-to-background ratios. Imaging was performed in 2D and 3D acquisition modes, with and without attenuation correction, on a modern PET scanner. Lesion detectability was visually and quantitatively assessed. For objects larger than 9 mm in diameter, target-to-background ratios ranging from approximately 3:1 to approximately 10:1 were detectable. Objects < 9 mm in diameter required a target-to-background ratio of >/=18:1. Target-to-background ratios required for lesion detectability were equivalent for 2D and 3D PET images with and without attenuation correction. In conclusion, 2D and 3D PET with attenuation correction consistently detected "tumors" >/= 9 mm. Lesions < 9 mm could be detected if there was high enough tumor uptake. No statistically significant differences in lesion detection were found for 2D versus 3D PET, or for attenuation-corrected versus non-attenuation-corrected images.