Scintillation detectors developed for PET traditionally use relatively thick crystals coupled to photomultiplier tubes. To ensure good efficiency the crystals typically measure between 10 and 30 nun thick. Detectors also require good spatial resolution so the scintillator is normally made up of a densely packed array of long thin crystals. In this paper, we present a novel design in which the detection crystal is divided into a number of layers along its length with an avalanche photo diode (APD) inserted between each layer. With thin layers of crystal, it is possible to use a continuous rather than a pixelated crystal. The potential advantages of this design over a conventional PMT-based detector are: (i) improved light collection efficiency, (ii) reduced dependency on dense crystal to achieve good stopping power, (iii) ease of crystal manufacture, (iv) reduced detector dead-time and increased count rate, and (v) inherent depth of interaction. We have built a four-layer detector to test this design concept using Hamamatsu S8550 APD arrays and LYSO crystals. We used the centre 16 pixels of each of the arrays to give an active area of 9.5 mm x 9.5 mm. Four crystals 9.5 mm x 9.5 mm were used with thickness increasing from 2 mm at the front to 2.5 mm, 3.1 mm and 4.3 mm, at the back, to ensure a similar count rate in each layer. Calculations for the thickness of the four layers were initially made using the linear attenuation coefficient for photons at 511 keV of LYSO. Experimental results and further simulation demonstrated that a correction to the thickness of each layer should be considered to take into account the scattered events. The energy resolution for each of the layers at 511 keV was around 15%, coincidence-timing resolution was 2.2 ns and the special resolution was less than 2 mm using a statistical-based positioning algorithm.