Given sufficient exposure to stimulus presentation, the oculomotor system generates a representation of the stimulus characteristics, which is then used to predict the upcoming target motion. In addition to compensating for the perceptual-motor delay, these predictive processes perpetuate eye motion during a transient occlusion and compensate for the loss of visual input. At present, however, it is not well understood whether and how the oculomotor system extracts and represents target acceleration for subsequent predictive control. To this end, we used a target occlusion paradigm where both position and velocity of the target during the occlusion and at reappearance could not be predicted without extracting target acceleration before target disappearance. We found that the oculomotor response during the blanking period was not influenced by target acceleration when the initial exposure was 200 ms. However, smooth and saccadic eye movements did discriminate between the different levels of acceleration after an initial 500- or 800-ms exposure. In the event that the smooth response during the occlusion did not match well the target trajectory and thus eliminate a developing displacement error, there was an increased saccadic displacement. Still, the combined response during the blanking period did not eliminate retinal slip and position error at target reappearance. These results indicate that information on target acceleration can be extracted on-line, during pursuit of a visible ramp, and then used to drive a predictive oculomotor response in the absence of visual input.