This in vitro study examines the effects of three preparation designs and different luting agents on the marginal integrity of partial ceramic crowns. One hundred forty-four extracted human molars were prepared according to the following preparation designs: A. Coverage of functional cusps, B. Horizontal reduction of functional cusps and C. Complete reduction of functional cusps. Partial ceramic crowns (Vita Mark II, Cerec 3 System) were bonded to the cavities with: Variolink II/Excite (Vivadent), Panavia F/ED primer (Kuraray), Dyract/Prime and Bond NT (Detrey/Dentsply), and Fuji Plus/GC cavity conditioner (GC). The specimens were exposed to thermocycling and mechanical loading. Marginal adaptation was assessed on replicas using quantitative margin analysis in the scanning electron microscope (SEM). Significant differences were observed between the preparation designs in general. Coverage of functional cusps with preparation of butt joints and use of Variolink as luting material showed a tendency toward the lowest values for compromised adhesion, especially within the dentin. Significant differences could be determined between luting systems: resin-modified glass ionomer cement (RMGIC) caused fracture of the restorations and revealed higher values than all other luting materials for compromised adhesion at ceramic-luting agent and tooth-luting agent interfaces. The dentin-luting material interface, in general, showed higher percentages of compromised adhesion (38-100%) than enamel- and ceramic-luting material interfaces (0-30%). In conclusion, the SEM data indicate that, with adhesively bonded partial ceramic crowns, retentive preparation is not contraindicated and the choice of luting material is more relevant than the preparation design. Margins below the cemento-enamel junction reveal significant loss of adhesion in spite of adhesive luting techniques. The RMGIC cannot be recommended as a luting material for feldspathic partial ceramic crowns.