This dissertation focuses on the problems existing in current system safety analysis,especially,modeling and solution of safety system,and integration of methods of safety analysis.There are still so many difficulties existing in the description and analysis of dynamic systems in current safety analysis techniques. In this case,we study the description and analysis idea based on classical Pctri Nets and its extension for safety systems,and present the way to model such systems using Generalized Stochastic Petri Nets (GSPN). For qualitative solution of the models,we define Critical States and bring forward an algorithm to work out the critical states of a safety system,so the path leading to high risk ha ard state can be eliminated with inner lock and time constraints,two of our new concepts. For quantitative solution,we study GSPN and find that the reachability graph of a GSPN is isomorphic to a Continuous Time Markov Process (CTMP) when the vanishing markings are eliminated,so Markov Process method can be used. And then we bring forward an approach to get the probability of steady-state based on Gauss-Seidel method and dynamic row-column techniques For the application of simulation in safety' analysis,sampling techniques for sparse events are discussed,and importance sampling method is brought forward to increase sampling efficiency. For integration of methods of safety analysis,we study and bring forward an approach to combine Petri Nets based method with other traditional safety analysis methods in order to make use of the advantages of the two. At last,a design of System Safety Modeling And Analysis Software Based On GSPN is provided,and a case study on a railway system using this prototype software has been carried out.