In this paper the use of neural networks for the identification of underwater vehicle dynamics is studied. A flexible way of identifying dynamics is desirable for several reasons. The dynamics of underwater craft are highly non-linear and cross coupling between various degrees of freedom normally exists. To date at best empirical models are available to describe these phenomena. On top of this the underwater environment can change drastically as a result of e.g. weather conditions. Due to their ability to adapt for changing circumstances in an online fashion, neural networks offer an interesting alternative for more conventional means of identification. This paper details an identification process using neural networks. To illustrate the performance of this identification process, these neural networks are then used directly or indirectly in a feedforward loop to control the craft in a simulation study.