In 3D PET measurements, activity outside of the direct field-of-view (FOV) is known to degrade signal-to-noise within the direct FOV, primarily by increasing the overall rate of random coincidences. In 3D brain studies, additional shielding around the patient has been used to address this issue. The additional shielding limits the acceptance of single-photons arising from regions just above and below the direct FOV, the singles FOV, and thereby reduces the randoms measured by the tomograph. In this work, we extend this idea for use in the torso with a shielding configuration consisting of two "clam-shell" shields of lead surrounding the patient both above and below the axial FOV. The lead is 6 mm thick by 10 or 20 cm in axial length, and curved into a C-shape to fit around the phantom or the patient's torso. The top half of the clam-shell rests on plastic wheels which travel on rails mounted to the edge of the patient pallet, and the separate lower half of each shield rests on a Styrofoam support beneath the bed. The shields are placed just above and below the direct field of view and as close to the patient as possible. Noise Equivalent Count (NEC) curves were measured with and without shields in place around an axially long cylindrical phantom. The NEC rates from 2D and 3D patient studies were calculated and compared with the phantom-derived NEC curves to determine the effectiveness of the axial shields. We find that the additional shielding offers a small improvement in phantom NEC at the high end of the nominal activity range for whole body FDG studies. 3D acquisition in two patients produces an NEC advantage of approximately a factor of two. With the addition of an axial shield, there may be a small additional improvement at the high end of this activity range.