his paper summarizes our experience in using a component model for constructing agents for agent-based simulations. In this model, named MALEVA, components encapsulate various units of agent behaviors or activities (e.g., follow gradient, flee, die, reproduce). Among its specificities, it extends the principles of software composition to the specification of control, through the notions of control ports and control components. A notion of composite component allows complex behaviors to be constructed from simpler ones. Some examples illustrate how our model may support a progressive construction of agent behaviors and also various forms of potential reuse. We also discuss the benefits of our model for a fine grain control of activation and scheduling.