Fast (up to 30 degrees /s) anticipatory smooth pursuit eye movements can be built up with repeated transient motion stimuli. It is thought that such stimuli charge a putative internal store of velocity information that can then drive anticipatory movements in the absence of a target. The aim of this study was to investigate the longevity of this store. Previous experiments with single ramp stimuli (Wells and Barnes 1998) suggested that the store lasts for only a few seconds before decaying to a baseline level. In the current study we investigate the possibility that the store was not maximally charged by single stimuli, precipitating its decay. The magnitude of the anticipatory response was indexed by smooth eye velocity 100 ms after target onset ( V(100)). In experiment 1 the build-up of the anticipatory response was examined by presenting sets of stimuli (comprising from one to five ramps) within a tracking phase and leaving a dark period (the 'gap') of 9.6 s between successive tracking phases. Each ramp was preceded by an audio warning cue and was accompanied throughout its 480 ms duration by an audio tone. Audio cues continued during the gap to reinforce timing information. V(100) for the first and last ramps of each set increased as the number of ramps was increased from one to three but reached an asymptotic level thereafter, suggesting that the velocity store is maximally charged after three presentations. In experiment 2 the store was maximally charged by presenting five ramps in each tracking phase and its decay was examined by leaving gaps of either 7.2 s or 14.4 s between successive tracking phases. V(100) was not diminished after either gap interval. In experiment 3 the velocity store was less well consolidated during tracking phases comprising two ramps. V(100) for the first response after the gap was unaffected by the 7.2 s gap interval but was significantly reduced when the gap interval was 14.4 s. The interval between the warning cue and ramp onset strongly influenced the magnitude of the anticipatory response, the optimum level being elicited by a cue time of 600 ms. In conclusion, this study has shown that the internal velocity store can be sustained for periods as long as 14.4 s provided that it is initially charged to a sufficiently high level and that accurate external timing cues are provided. Furthermore, we provide evidence to suggest that this process may be controlled by a two-part sample and hold mechanism.