Chains of impact craters, or catenae, have been identified in Voyager images of Callisto and Ganymede. Although these resemble in some respects secondary crater chains, the source craters and basins for the catenae cannot be identified. The best explanation, proposed by Melosh and Schenk, is a phenomenon similar to that displayed by former comet P/Shoemaker-Levy 9; tidal (or other) breakup close to Jupiter followed by gradual orbital separation of the fragments and collision with a Galilean satellite on the outbound leg of the trajectory. Because the trajectories must pass close to Jupiter, this constrains the impact geometry (velocity and impact angle) of the individual fragments. For the dominant classes of impactors, short-period Jupiter-family comets and asteroids, velocities at Callisto and Ganymede are dominated by jovian gravity and a satellite's orbital motion, and are insensitive to the pre-fragmentation heliocentric velocity; velocities are insensitive to satellite gravity for all impactor classes. Complex crater shapes on Callisto and Ganymede are determined from Voyager images and Schmidt-Holsapple scaling is used to back out individual fragment masses. We find that comet fragment radii are generally less than ~500 m (for ice densities), but can be larger. These estimates can be compared with those for the Shoemaker-Levy 9 impactors. ¿ American Geophysical Union 1995