This thesis is dedicated to the study of electrical power system security analysis in steady state. The problem consists in identifying quickly and precisely the contingencies which create thermal constraints violations. The suggested approach is based on the use of artificial intelligence techniques and specifically on expert systems. Who says expert system, says knowledge. This thesis presents different approaches for knowledge acquisition. It is proposed to exploit quantitative as well as qualitative knowledge. An approach based on the systematic analysis of single contingencies is advanced to establish a quantitative knowledge base. The method consists in quantifying the effect of each branch on the other branches of the power system. The knowledge base can be used in two different ways. On the one hand, it allows the estimation of the power flows in a branch following a single contingency around the working point for which it has ken established. This first use could be also extended to the analysis of double contingencies by considering the contingencies for which the superposition principle can be applied. On the other hand, this same knowledge base could be used to automatically extract rules for a contingency. Rules are extracted by considering similarities between the knowledge bases obtained for different characteristic operating states. This thesis also deals with knowledge obtained by experienced power system operators. The rules are expressed by considering the main factors which allow a qualitative identification of the critical contingencies. An approach based on fuzzy filtering is recommended to process the knowledge. Furthermore, this work presents a study of contingency ranking methods by order of severity. The methods performances are analyzed with the intention of integrating them within an expert system. The different knowledge forms collected and developed have allowed the development of an expert system prototype. The prototype performances have been evaluated by considering the Swiss utility system. The proposed approach is greatly time-saving in comparison with a classical approach based on the use of a power flow calculation, while maintaining an acceptable results precision.