This study quantifies the different scatter components in PET and examines how the different components degrade reconstructed PET images. METHODS We simulated the measurement of various phantoms using Monte Carlo (MC) calculations and compared the MC-generated projections and images with the corresponding experimental data. The coincidences were subdivided in four classes: primaries, object scatter (scattered in the object only), gantry scatter (scattered in the scanner only) and mixed scatter (scattered both in the object and the scanner). RESULTS In the projections of the line sources, the gantry scatter was closely located around the source position, whereas the object scatter was smeared over the whole field of view and could be parameterized well by a monoexponential function. The mixed scatter had nearly the same distribution as the object scatter, but with a smaller amplitude. The calculations and experimental data were in excellent agreement; i.e., led to the same parameterization of the scatter distribution functions and to a similar localization of the scatter components in the reconstructed images. CONCLUSIONS The spatial distribution of the scatter components justifies the widely-used assumption that it is sufficient to restrict experimental scatter correction techniques to the object scatter. Furthermore, it is possible to derive the parameters for the scatter kernels, which are needed for the convolution-subtraction algorithm, by MC simulations.