DivX and other MPEG formats compress movies using keyframes and reference frames. Keyframes are self-contained compressed frames, like JPEG images. Reference frames then only contain information that has changed compared to the previous frame.

So, in order to skip/forward or rewind, you need to jump to a keyframe (since they're self-contained). If you jump to a non-keyframe, it could not look at that frame's data to figure out the entire picture. An application would either skip to the nearest keyframe, or read the previous keyframe and attempt to construct the current frame by looking through all the reference frames in between.

Some people encode movies with only 1 keyframe... the very first frame. If you ever try to skip to another frame, it has to jump back to the beginning. Any application that would rebuild the video file for rewind/fast forward would insert new keyframes throughout the video so you can skip to them. The idea of using just 1 keyframe is to reduce the file size (each keyframe adds Kbs).

Some Applications like DivFix or VirtualDUB can sometimes be used to fix these issues. However these applications are beyond our scope.