The ultimate function of the nervous system is to produce behaviours that enable animals to survive and reproduce. Accordingly, if we want to understand how the brain works it is crucial to take into account its final output. Ethology postulates that complex behaviours are formed by sequences of simpler units of movements. The identification and characterization of these behavioural motifs could be useful to understand how the neuronal circuits that underlay them work. We begin, in chapter 1, by discussing the advantages and disadvantages that exist in using supervised and unsupervised methods to classify behavioural motifs, and the radically different solutions that both approaches can produce. In chapter 2 we use a simple supervised method based on kinematic parameters of bouts to classify the movement types that zebrafish larvae execute while moving at different speeds, and confirm that, like other vertebrates, larvae use two gaits to control speed. The third chapter reports a new general purpose clustering algorithm based on the valley between density peaks that is able to automatically detect the number of clusters on a wide variety of synthetic and real live data sets. In chapter 4 we describe a novel automatic tracking system and how we used it to acquire a large collection of movements that zebrafish larva preformed while engaged in a wide range of behaviours. In chapter 5 we apply density valley clustering to this collection of larval movements and design an approach to compare similarity between clusters of movements. We found, free from human supervision, eleven bout types, seven of which had been described previously and four that are novel. Finally, in chapter 6, we use this classification to describe the sequences of bouts that larvae form while responding to continuous stimuli. By using the transitions between bout types we create a space where the sequences that fish execute in response to particular stimuli cluster into behavioral states.