This paper presents an overview of crowd simulation models, their limitations, and an alternative agent-based approch. First we introduce several methods and then we focus on two widely used and validated simulation tools that use grid-based models. We discus the artifacts that these models introduce regarding the way they treat the space and the implication that this has in the movement of the agents during the simulation. We also describe the limitations that current commercial software tools have in terms of simulating human psychology and physiology. The paper discusses an agent-based alternative approach developed to overcome these limitations. The model allows for the simulation of human movement that can provide results more closely describing behavior of real people during an emergency situation. Flow rates, densities and speeds emerge in our model from the physical interactions between people instead of being predefined.