A more formal definition: A decision problem C is Co-NP-complete if it is in Co-NP and if every problem in Co-NP is many-one reducible to it. This means that for every Co-NP problem L, there exists a polynomial time algorithm which can transform any instance of L into an instance of C with the same truth value. As a consequence, if we had a polynomial time algorithm for C, we could solve all Co-NP problems in polynomial time.
Each Co-NP-Complete problem is the complement of an NP-complete problem. The two sets are either equal or disjoint. The latter is thought more likely, but this is not known. See Co-NP and NP-complete for more details.