The diversity of plants we investigate has led us to define a generic methodology for measuring real plants and modelling their growth. This methodology is implemented in the AMAPmod software, which provides the user with various tools for encoding, exploring and modelling plants. In a typical experiment, botanical structures of plants are measured in the field, possibly at different scales of detail. These structures are encoded using a dedicated language in a coding file. From this description, a formal representation of the plant structures can be built by the AMAPmod software. This model of the plant structures can represent several plants, both at different scales (e.g. internode scale and branching system scale) and at different dates. The AMAPmod software provides a natural feedback on the measured plants by creating realistic 3D graphical reconstructions. The user can then interact with this formal representation of real plants using the AMAPmod querying language. Various types of data can thus be extracted at different levels of complexity. From the extracted samples of data, models belonging to different families can be estimated. Some of these models are used as analysis tools in order to reveal levels of organisation which cannot be studied efficiently using only direct observations. Such a methodology introduces new perspectives in agronomy by giving a central role to the structure of plants. (Resume d'auteur)