This paper relates an experience in using a component model to design and construct agents. After discussing various rationales and architectural styles for decomposing an agent architecture, we describe a component-based model of agents, named MALEVA. In this model, components encapsulate various units of agent behaviors (e.g., follow gradient, flee, reproduce. . . ). Among its specificities, it provides an explicit notion of control flow between components, for a fine grain control of activation and scheduling. Moreover, a notion of composite component allows complex behaviors to be constructed from simpler ones. Two case studies in the domain of multi-agent based simulation are presented in this paper. They illustrate the ability of the model to support both a bottom-up and a top-down approach for agent architecture design.