Humans readily make predictive saccades to periodic alternating targets. This predictive behavior depends on internal monitoring of timing error of past saccades in order to determine the time of initiation of future saccades; our earlier studies have confirmed this by finding correlations between latencies of consecutive predictive saccades. It is natural to consider that timing error is determined by visual detection of the difference between the time the target appears and the time the eyes arrive at the target; this in turn implies that saccades must actually be produced in order for their timing errors to be determined and predictive saccade timing to be established. We tested this hypothesis by having subjects view alternating visual targets while fixating a central target in order to eliminate saccade production. After six alternating target presentations, subjects began tracking the alternating targets. Tracking performance was assessed with an error measure that compared saccade latency and inter-saccade interval with desired values (zero and inter-stimulus interval, respectively). Errors in this Prior Viewing paradigm were compared to those from a conventional De Novo paradigm in which saccades began as soon as the alternating targets were presented. Saccades under Prior Viewing reached a low-error steady predictive state more rapidly than under De Novo tracking. The initial saccade under Prior Viewing had a higher latency than the others, suggesting that this saccade was reactive even though the paradigm is predictable; other reasons for this higher latency include time to disengage from the fixation target and time required to pre-program the initial set of saccades. The results show that visual detection of timing error from an actual motor act (saccades) is not necessary to establish predictive saccadic pacing: sensory-only information from viewing the moving targets can help to establish this predictive state.