To evaluate the cavity sealing obtained after thermocycling with five adhesive systems in which one all-in-one adhesive was compared to three one-bottle adhesives, and to observe the effect of a low-charged resin layer added to a one-bottle adhesive. Twenty-five recently extracted teeth were randomly allocated to five experimental adhesive systems (n = 5 each): Optibond Solo (OS), Scotchbond 1 (SB1), PQ 1, Prompt-L-Pop (PLP), SB1+Revolution (R). On each tooth, two rectangular cavities at the cementoenamel junction were filled with a microhybrid composite (Z100) and the tested adhesives. Teeth were thermocycled and stained with AgNO3 + vitamin C. Leakage was evaluated on a 6-point (0-5) severity scale and the highest score was selected for each restoration. Results were treated by ordinal logistic regression and considered to be significant at p < 0.05. No significant difference was found between leakage values for enamel and dentin interfaces. Leakage scores never exceeded 2 for OS, SB1 and SB1+R, while they reached a maximum of 5 in 20% of PLP cases. OS was significantly better than the other adhesive systems, which were statistically equivalent. The addition of a flowable composite layer on SB1 did not yield a significant difference, but tended to give better results mainly at the dentin interface. The all-in-one adhesive PLP, because of quite variable results, seems less reliable than the one-bottle adhesives, of which OS provides significantly the best results. Addition of a flowable composite on SB1 appears to yield slightly better results.