The aim of this paper is to report on experience with 3D positron emission tomography (PET) in our institute where we have three 3D scanners, of which two operate exclusively in 3D mode (ECAT ART, EXACT 3D). Fully 3D PET requires attention to a number of factors which are not as problematic in 2D PET. Firstly, 3D tomographs designed for whole-body acquisition suffer from a large single-photon field of view, extending well beyond the coincidence field of view. Single photons from outside the coincidence field of view increase the dead time and random coincidence rates, and contribute scattered events. For brain studies, we have extended the lead side shielding at the ends of the tomograph to mitigate against these effects, and this has dramatically improved the count rate performance. This approach is not as effective for whole-body scanning. In addition, operating in 3D without septa necessitates new approaches to transmission scanning, as measurements using positron emitters such as 68Ge/68Ga have the unfavourable characteristics of high dead time and high scatter. Both of our fully 3D scanners use 137Cs for single-photon transmission measurements, although the data are treated differently. On the ECAT ART, a combination of physical and electronic collimation effectively reduces transmission scatter to acceptable levels. On the EXACT 3D physical collimation is not as readily implemented and therefore segmentation and reassignment of the histogrammed attenuation (mu) values is employed to produce unbiased attenuation correction factors in 3D. Many of the lessons learnt with these BGO (bismuth germanate) based tomographs will be applicable to the next generation of systems using faster detectors such as lutetium oxyorthosilicate (LSO).