A routing loop occurs when a distributed path selection mechanism creates a (usually temporary) loop in the path to a particular destination; i.e. the routing table entries in a group of switching nodes, with respect to given destination, form a closed loop. Any traffic for that destination that enters that loop never leaves - just loops endlessly until the loop is broken.
Routing loops generally form as a result of out-dated data being used by some of the switching nodes in the loop, when doing their path selection computations - e.g. after another switching node has failed, and paths which previously went through it are being re-computed. They generally do not last long, and disappear as updated data arrives.