They are implemented in very different ways.
hash_map (unordered_map in TR1 and Boost; use those instead) use a hash table where the key is hashed to a slot in the table and the value is stored in a list tied to that key.
map is implemented as a balanced binary search tree (usually a red/black tree).
An unordered_map should give slightly better performance for accessing known elements of the collection, but a map will have additional useful characteristics (e.g. it is stored in sorted order, which allows traversal from start to finish). unordered_map will be faster on insert and delete than a map.