Positron emission mammography (PEM) is a new, specialized imaging modality utilizing PET radiopharmaceuticals to detect breast cancer. The capabilities and limitations of PEM in detecting breast tumors were investigated with a series of phantom experiments. The PEM imager was mounted on a standard Lorad biopsy table (separated by 18 cm). In the initial phase of the investigation, basic scanner parameters (resolution, sensitivity, and scatter fraction) were measured. The effects of a number of breast imaging parameters (length of acquisition, breast thickness, and breast density) on detection of breast lesions were then explored utilizing special phantoms. Moderately compressed breasts were simulated with a block of gelatin containing amounts of FDG consistent with 370 MBq injections. Lesions were simulated with four hollow spheres (inner diameters=5 mm, 8 mm, 12 mm, and 15 mm) filled with amounts of FDG representative of uptake in malignant breast tumors (target-to-background concentration ratio=8.5:1). Resolution at the center of the imager was 3.9 mm, sensitivity was 0.059 kcps/kBq/ml and the Compton scatter fraction was approximately 12%. Objects as small as 8 mm in diameter could be detected after 30 s of data acquisition; 5 mm spheres were detectable after 300 s. Object detection capabilities were reduced with increasing breast thickness. In thin compressed breasts (2 cm) even the smallest sphere (5 mm in diameter) could be detected; increasing breast thickness increased the minimum detectable sphere diameter to 8 mm. Increased background activity caused by FDG uptake in metabolically active normal tissue more prevalent in radiodense breasts compared to "fatty" breasts was simulated and shown to reduce the minimum detectable lesion size to 12 mm for the densest breasts. These results demonstrate the potential of PEM for the detection of breast lesions. The addition of the system to a standard biopsy apparatus indicates its potential for use to guide some core biopsies of breast cancers.