We have used noise-equivalent count (NEC) rates to optimize count rate performance for 3D acquisition in PET in a wide range of situations, with particular reference to imaging of the torso. We have also compared NEC performance for 2D and 3D acquisition in order to establish the conditions under which 3D mode offers an improvement over 2D mode. Measurements were performed on four tissue-equivalent phantoms ranging in size from that of an infant's head (13 cm diameter) to that of an obese adult's chest (37 cm x 48 cm). Count rate data were acquired as a function of phantom size, activity in the field of view, lower energy discriminator level (LLD) and acquisition mode, and NEC rates were derived as a function of these variables. The LLD at which the highest NEC rate is obtained shows a dependence both on phantom size and on the activity in the field of view both for 2D and for 3D acquisition. The relative advantage of 3D mode over 2D mode, at the optimum LLD setting, is also strongly dependent both on activity in the field of view (FOV) and on the phantom size. The main limiting factors for 3D NEC rates are detector dead-time for small phantoms and random coincidences for large phantoms. The 3D NEC rate is more than twice as great as the 2D NEC rate when less than 60 MBq is present in the FOV for all phantoms except the largest, in which case a ratio of two is only achieved for activities less than 25 MBq. For the smallest phantom, 3D/2D NEC ratios of greater than 3.5 are obtained when the activity in the FOV falls below 10 MBq.