This paper describes the constitution and the statistical exploration of a plant architecture database using the AMAPmod system. In AMAPmod, plant architectures are represented by a formal model, called MTG, which is the central data structure of the system. A dedicated querying language AML enables the user to access plant architecture data with efficient built-in primitives. By combining these primitives, the user can extract various types of data that preserve a more or less important part of the structural information of the plant. Specific statistical tools have been designed to analyse data samples extracted from the plant architectures. Most of these tools, ranging from (hidden) Markovian models to dynamic programming comparison techniques, apply to samples of sequences. AMAPmod methodology is illustrated on an actual-scale example concerning a hybrid family of apple trees.