Structure was imposed on a tapping task by requiring participants to reproduce sequences of responses to icons organised in spatial clusters. A first experiment featured sequences either segregated or not segregated by clusters. Accuracy was higher for sequences segregated by clusters. Moreover, inter-response times were longer at cluster boundaries than within cluster boundaries. To rule out possible confounding effects of movement length, this temporal pattern was replicated in a second experiment requiring a single response indicating the next sequential step, following the presentation of a portion of a previously practised sequence. These results suggest that sequence reproduction can be sustained by a hierarchical representation based on spatial proximity and provide a first indication of the role of spatial structure in serial-spatial memory.