In this thesis, we study continuum tree-valued processes. First, we define an abstract framework for these processes, by constructing a metric on the space of locally compact, complete R-trees, endowed with a locally finite Borel measure. This topology, called Gromov-Hausdorff-Prokhorov topology, allows for the definition of tree-valued Markov processes. We then give a new construction of the pruning process of Abraham-Delmas-Voisin, which is an example of a Lévy tree-valued process. Our construction reveals a new genealogical structure of Lévy trees. Furthermore, it is a path wise construction, which describes the transitions of the process explicitly. We apply this description to the study of certain stopping times, such as the first moment the process crosses a given height. We describe the process at that time through a new spinal decomposition. Finally, we focus on the Aldous-Pitman fragmentation of Aldous's Brownian tree. Following Abraham and Delmas, we study the effect of the fragmentation on discrete subtrees of the Brownian tree. The number of cuts needed to isolate the root, suitably renormalized, converges towards a Rayleigh-distributed random variable; we prove a Central Limit Theorem describing the fluctuations around this limit
