Abstract : Multi-Agent simulations allow to take into account the way the entities of a system interact one with each other, and therefore allow to link the microscopic and macroscopic levels. Financial markets are often studied with the help of group centered models, which exhibit their limits when one searches to explain the emergence of sorne phenomenoms observable in price series. We therefore propose a model of financial market centered on the behavior of its components, which allows to reproduce realisticly their functioning both at an intraday and extraday scale. This model first allowed us to propose a theory to explain the origin of stylized facts, which are statistical properties observable in all price series, on all types of financial markets. Our proposition shows that stylized facts seem to be merely caused by the way the market is organized and the way economic agents interact through it. ln a second part, we used our model to study extreme events that can be found in price series which are named "bubbles" and "crashes". We have illustrated, with the help of our model that these critical episodes can happen when a sufficient part of investors adopt a speculative investment strategy which is coherent with sorne theories proposed by economists.