The aim of this in vitro study was to evaluate dentin adhesion after cementation of immediate direct "All Purpose" Hybrid (AP.H) composite inlays (Dentsply) and Cerec Dicor-MGC (Dentsply) inlays with the dentin adhesive Prisma Universal Bond 2 (Dentsply) and the dual-curing Dicor-MGC luting composite (Dentsply). In 24 extracted human molars, standard MOD cavities were prepared with one approximal margin located in enamel and the other one located in dentin. They were divided into four groups: (I) AP.H inlays, luting composite only, chemically cured; (II) AP.H inlays, luting composite, immediately-light-cured; (III) AP.H inlays, luting composite, initially chemically and delayed-light-cured (15 min); and (IV) MGC inlays, luting composite, initially chemically and delayed-light-cured (15 min). In vitro load cycles corresponding to five years of clinical stress followed. Initially and after specimens were loaded, the margins were analyzed quantitatively by SEM. The tooth/cement and cement/inlay interfaces were scored separately. The initial percentages of "continuous margin"--at both the tooth/luting composite and luting composite/inlay interfaces--were higher than 94% for all groups. At the end of the load cycles, the quality of the margins at the tooth/luting composite interface significantly decreased for all groups. The highest decrease was found for the cervical margins located in dentin, where only 37%-61% were scored as "continuous margin". The AP.H inlay/luting composite interface showed almost no change. At the MGC inlay/luting composite interface, the percentage of "continuous margin" decreased to 74%. After specimens were loaded, the percentage of "continuous margin" in dentin was lower than in enamel, despite the use of a dentin bonding agent