DNA end-joining is the major repair pathway for double-strand breaks (DSBs) in higher eukaryotes. To understand how DSB structure affects the end-joining process in human cells, we have examined the in vivo repair of linearized plasmids containing complementary as well as several different configurations of non-complementary DNA ends. Our results demonstrate that, while complementary and blunt termini display comparable levels of error-free rejoining, end-joining fidelity is decreased to varying extents among mismatched non-complementary ends. End structure also influences the kinetics of repair, accurately recircularized substrates for blunt and complementary termini being detected significantly earlier than for mismatched non-complementary ends. These results suggest that the end-joining process is composed of an early component, capable of efficiently repairing substrates requiring a single ligation event, and a late component, involved in the rejoining of complex substrates requiring multiple processing steps. Finally, these two types of repair events may have different genetic requirements as suggested by the finding that exposure of cells to wortmannin, a potent inhibitor of phosphatidylinositol 3-related kinases (PI 3-related kinases), blocks the repair of complex substrates while having little or no effect on those requiring a simple ligation event.