We have developed an exposed brain preparation for observing growth cone pathfinding behavior while performing in vivo pharmacological manipulations, and we used it to test whether Xenopus retinal growth cones need filopodia to navigate. Time-lapse video observation showed that cytochalasin B acted quickly and reversibly when applied; cytochalasin B-treated growth cones lacked filopodia, but had active lamellipodia and continued to advance slowly. Whereas normal retinotectal axons visualized with horseradish peroxidase turn caudally in the mid-diencephalon to reach the tectum, cytochalasin B-treated axons grew past the normal turning point and, instead, continued straight within the diencephalon. In dose-response experiments, pathfinding became abnormal in the same concentration range in which filopodia disappeared. These results suggest that filopodia are necessary for retinal growth cones to respond to guidance signals in the diencephalon.