The reproduction, in complete darkness, of sequences of 5 horizontal saccades towards previously presented visual targets has been investigated in human subjects on the ground (control subjects) and one cosmonaut in microgravity. The incidence of corrective saccades during the execution of the memory-guided saccades in darkness has been examined. It was quite large for the control subjects (more than half of all saccades), and increased during the flight, while the corrective visually guided saccades incidence decreased. Direction errors occurred in about the third of all sequences on the ground, and this parameter also increased in microgravity. Memory-guided sequences were mostly hypermetric. Whereas the absolute error continuously increased with the target rank, it was not the case with the amplitude ratio, which presented a peak at the third rank, that is, at the middle of the sequence. The accuracy of the reproduction of the sequences did depend on the sequence pattern as much as on the subject. Some learning was observed in repeated reproduction of the same pattern. Although the average error did not change in microgravity, the linear regression coefficient between the visually guided and memory-guided saccades decreased.