Natural behavior must deal with inherent delays in visual feedback. This problem is particularly acute in dynamic environments such as playing sports. One solution is to learn environmental dynamics to predict future states, similar to proposals for predicting sensory consequences of movements (e.g. Wolpert et al., 1998). We present evidence for predictive control of eye movements of two skilled squash players. Eye movements were recorded simultaneously from both subjects using two RIT lightweight wearable eye-trackers (Babcock & Pelz, 2004). Gaze was analyzed when subjects observed the ball after the opponent hit it (receiving), and after the subject hit the ball back to the other player (returning). When receiving, Player 1 fixated the front wall 110ms before the ball arrived with 8° of error on average. Player 2 fixated 200ms before arrival, with 13° of error. When returning, fixations began 31 (Player 1) and 57 ms (Player 2) after the bounce on the front wall. Players pursued the ball for 64% (Player 1) or 41% (Player 2) of its path from wall to the floor. Player 1 was able to pursue the ball up to speeds of 190°/s and at reduced gain up to 230°/s, while Player 2 pursued up to 117°/s, and to 160°/s with reduced gain. We conclude that the ability to make high-precision anticipatory fixations, and to pursue at such high speeds, is evidence for predictive mechanisms due to learnt models of the ball's dynamic properties. In addition, the degree of prediction depends on the subject's behavioral goals.