To perform Monte-Carlo simulations of PET tomographs, we used SimSET (a Simulation System for Emission Tomography). Since we need to estimate performance characteristics of different tomographs, this widely used simulation package had limited properties for our use. We integrated a module to generate random events from a singles simulation and we modeled dead time losses. The count rate model we used is based on Moisan & al's method (1997) developed for the Siemens HR+ scanner. We needed a reliable simulation tool, which could provide data such as scatter fraction, sensitivity, count rates or noise equivalent count rate (NEC). So, we carried out a two-steps validation of SimSET in 3D alone and with additional features. First, we compared results obtained with SimSET and with two other simulators (GePEToS, a PET simulator using Geant4 and a simulator developed by Adam & al. (1998)). We estimated energy spectrum and sensitivity to scattered events. Second, we compared our simulated data with experimental data obtained on two Siemens tomographs : the ECAT HR+ and the ECAT ACCEL. We determined scatter fraction, sensitivity (global and for each plane) and all the different count rates (true, scattered and random). We also calculated the NEC to completely validate this simulator. We found that SimSET and the features we integrated (random events and dead time losses) are rather in good agreement firstly with two other simulators and secondly with experimental data. Further validation is planned concerning spatial resolution and 2D mode.