We studied the performance of memory-guided sequences of saccades in 20 healthy subjects. Subjects were presented sequences of two, three, four, or five subsequent visual targets at a frequency of one target per second, which they had to perform after a delay of three seconds. Thus, the temporal order of the presented targets had to be memorized in this paradigm. The mean percentage of correctly performed sequences of saccades was 97%, 86%, 52%, and 32% for two, three, four, and five sequentially presented targets, respectively. The intersubject performance was variable, with some subjects not being able to perform quintuple memory-guided sequences at all. A multiple regression analysis revealed that greater numbers of clustered targets (i.e., targets closer than 2.50) led to an increased rate of erroneously performed memory-guided sequences of saccades. This study demonstrates the relatively small capacity of the working memory for chronological order, an aspect of working memory which was not studied systematically until now. Furthermore, there is a remarkable intersubject variability for longer sequences. Lastly, visuospatial properties of the target presentation such as the number of clustered targets are crucial for the performance of such saccade sequences, whereas neither the hemifield of the presentation of the targets nor the number of changes in the direction of the saccades seem to influence the performance. These results have implications for clinical research using such sequences, since longer sequences and the effect of cluster presentation of the visual targets could bias the study.