Positron emission tomography (PET) is a nuclear medical imaging modality. Its aim is to visualize the 3-dimensional distribution of a radiopharmaceutical (also called the tracer) within a patient (clinical PET) or test-animal (in case of preclinical investigations). The information that can be obtained from the reconstructed distribution depends on the specifics of the tracer. In general it is linked to a certain physiological processes. The most common clinical application of PET makes use of (18F-)FDG: a 18F-labeled, glucose-like molecule that targets the glucose metabolism. Consequently, an FDG PET scan visualizes the glucose uptake of different tissue types, has proven to be a very valuable tool in the diagnosis, staging, and monitoring of carcinoma and metastasis. In PET the tracer molecules are labeled with position emitting isotopes. Positrons annihilate close to their point of emission with electrons that are abundant in the surrounding tissue. As this takes place, a pair of 511 keV ?-photons is emitted on (almost) antiparallel trajectories. The detection of both photons belonging to one such annihilation pair allows localizing their point of origin to a so-called line-of-response (LOR), which is defined between the two points of detection. The decision if two detected ?-photons originate from the same annihilation event is based on the difference in their detection times, i.e. two ?-photons are classified as an annihilation pair if they are detected within a small time window. Depending on the count rate within the individual detector elements and the width of the coincidence acceptance window two uncorrelated ?-detections might be misclassified as annihilation pair. These events are referred to as random events or randoms. Detecting a large number of annihilation photon pairs facilitates the 3d reconstruction of the PET-tracer distribution. The practical, clinical value of a reconstructed PET image crucially depends on how accurately the original tracer distribution can be reconstructed from the acquired PET data. This accuracy is fundamentally limited by the number of detected (true) annihilation photon pairs, the ratio between true, random and scattered events, as well as the accuracy with which each ?-photon can be localized in space and time. All these limiting factors are largely determined by properties of the employed scintillator detectors: namely the spatial resolution, their timing resolution, their energy resolution, their ?-capture efficiency, and their depth of interaction (DOI) resolution. Consequently, an improvement of the detector performance in these key areas will lead to improved image quality, allow for shorter scan times per patient, and/or permit to work with reduced tracer activity. It is the aim of this work to investigate as to how far the performance of detectors for application in PET can be improved by using new materials and innovative detector design. In this, a particular emphasis will be on the timing performance of the detectors as it has been shown that utilizing so called time-of-flight information can lead to a drastic reduction of statistical noise in the reconstructed image which is often the dominating noise source in clinical PET. A key component in every scintillator detector is the photosensor that is applied to detect the scintillation light. A novel type of photosensor that has received growing interest as a replacement for conventional photomultiplier tubes (PMTs) in TOF-PET detectors are so-called silicon photomultipliers (SiPMs). SiPMs are solid state light sensors that offer comparable gain and timing response with respect to PMTs but offer some distinct advantages. SiPMs are insensitive to magnetic fields and can be produced to be nonmagnetic, which allows for integration in magnetic resonance imaging (MRI) devices. Furthermore, SiPMs are essentially transparent for 511 keV ?-photons, they are compact and rugged and can be in integrated relatively easily into pixel-arrays, which allows for novel detector designs. In order to make the best use of the potential offered by of SiPMs it is important that subsequent front end electronics are optimized to meet the specific demands of these sensors. Such an optimization requires a detailed understanding of the properties of SiPMs as well as the generated electronic signals. Important insights in these matters can be gained from equivalent circuit models such as the one presented in chapter 2. In this chapter it is demonstrated how such a model can be applied to simulate the combined electronic response of the SiPM and subsequent front end electronics. These simulations constitute an invaluable tool for the development of suitable preamplifiers, they help to understand the signal generation and signal transport within SiPMs, and can predict unexpected effects such as the electronic signal nonlinearity. A further potential improvement of the TOF-PET detector performance can be achieved by replacing the scintillation material. A particular interesting candidate in this respect is the recently discovered LaBr3:Ce due to its high light yield (~70 photons / keV), its fast decay time (15 ns) and its good intrinsic energy resolution (< 3% at 511 keV). Especially the conjunction of high light yield and fast decay make LaBr3:Ce highly interesting for application in fields where accurate timing is essential. The outstanding timing capabilities of LaBr3:Ce in combination with SiPMs and an optimized readout architecture are demonstrated in chapter 3. The measurements described there yield an exceptional coincidence resolving time (CRT) of 100 ps (FWHM) for two detectors utilizing LaBr3:5%Ce crystals with a size of 3 mm × 3 mm × 5 mm. Still, the measurements presented in chapter 4 show, that the favorable timing properties of LaBr3:Ce are somewhat diminished by a much slower rise time of the scintillation pulse when compared to e.g. LYSO (280 ps up to 2 ns compared to 90 ps for LYSO). A slower buildup of the scintillation signal increases the statistical variation of the arrival times of individual scintillation photons at the photosensor. In consequence, the time stamp that can be obtained in combination with a specific photo detector degrades. This is shown and quantified by means of a comprehensive model that predicts the timing resolution of SiPM based scintillation detectors (chapter 5). The model predictions and the corresponding validation measurements show that the CRT that can be obtained with LYSO-SiPM-detectors (CRT = 138 ps FWHM) is nearly as good as for the LaBr3:5%Ce based detectors. The presented data and model predictions also reveal the importance of the careful optimization of the applied trigger scheme as an ill conditioned trigger can easily degrade the timing performance by more than 50 %. Further insight in the mechanism behind the surprisingly similar timing performance of LaBr3:5%Ce and LYSO is given by means of a rigorous statistical description of the photon detection process in chapter 6. As the generation/emission, transport, and detection of scintillation photons are independent statistical processes it is possible to calculate the lower limit on the timing resolution in a relatively simple formalism based on the timing information carried by each individual photon. The calculation of the so-called Cramér-Rao lower bound on the timing resolution associated with the detectors that were used in the above mentioned experiments shows that the observed CRTs of both types of detectors is almost entirely governed by the statistical properties of the photon detection indicating that little to no improvement can be expected from further optimization of readout electronics or time stamp generation. Lastly, the concept of monolithic scintillator detectors read out by pixelated sensor arrays is explored. Utilizing monolithic crystals instead of pixelated blocks offers several advantages such as a good special resolution while maintaining a good energy resolution, intrinsic DOI-information, ease of assembly, and faster light transport. In chapter 7 and chapter 8 two detector prototypes based on this concept are characterized. In both cases the detectors employ 10 mm thick monolithic scintillation crystals with footprints matching 4 × 4 SiPM arrays (13.2 mm × 13.2 mm and 18.0 mm × 16.2 mm, respectively). The detector described in chapter 7 is based on the first available SiPM array (SensL SPMArray 3035G16) and an LYSO:Ce scintillator. This detector exhibits a good energy resolution (?E/E = 14%) and spatial resolution (Rs < 1.6 mm in the detector center). Furthermore, the capability of the detector to correct for DIO effects was shown. Yet, the due to the low PDE and comparatively slow electronic response of the SiPMs the CRT measured for this detector was moderate (CRT = 1.4 ns FWHM). The second detector prototype, which is characterized in chapter 8, utilizes an SiPM array with much improved PDE and timing properties (Hamamatsu MPPC S11064–050P). The spectral sensitivity of this sensor array furthermore allowed for the use of LaBr3:5%Ce. The applied changes in the detector materials and read out architecture amounted to a dramatic improvement of the timing resolution to CRT = 200 ps (FWHM) and energy resolution (7.8 %) and a slight improvement of the spatial resolution (Rs < 1.5 mm in the center of the detector). The results presented in this work clearly show that monolithic scintillator detectors combined with SiPM readout can outperform conventional PMT-pixelated-block detectors in terms of spatial resolution, energy, and timing resolution. Nevertheless, the integration of monolithic scintillator detectors in TOF-PET and TOF-PET/MR systems requires further research and some important matters such as shortening the time consuming system calibration or the optimization of data handling and data processing still have to be addressed.