This paper aims at describing an Agent Based Model, SIGECORIS, designed to generate data on the interactions in the long run of heterogeneous stakeholders facing flood risk. Among all issues pointed out in the literature on flood risk management, two particular points were given a special focus in SIGECORIS: heterogeneity of granularity as well as of behavioural patterns of stakeholders involved on one hand, and flood considered rather as a constraint than as a driver of stakeholders` actions. Firstly, two types of heterogeneous stakeholders are implemented, investors (individual level) and communities (collective level). These stakeholders are heterogeneous in their role, their means, knowledge of each other actions and perception of flood risk. Secondly, neither of these two types of stakeholders do follow flood management as a primary objective. Flood risk is considered as a territory characteristic which constrains stakeholders` actions. Investors look at economic returns of their decisions under the constraints imposed by communities, while these look at the attractivity of their territory. This experiment can be considered as an example of implementing a companion modelling ([2,1]) for research purpose on flood management issue. We focus here on a first stage of the process: discussing the implications of this modelling for making explicit current assumptions in the field as well as the diversity of interactions among investors and communities.