Despite the fact that one-step adhesives are currently used routinely in clinical practice, long-term studies on their clinical performance are scarce. The objective of this randomized controlled clinical trial was to test the hypothesis that a 2-hydroxyethyl methacrylate (HEMA)-free one-step self-etch adhesive performs worse than a conventional multistep etch-and-rinse adhesive. Two-hundred and seventy-six noncarious cervical lesions in 52 patients were restored with a micro-hybrid composite (Gradia Direct; GC). These restorations were bonded in random order either with the HEMA-free one-step adhesive G-Bond (GC) or with the 'gold-standard' (control) three-step adhesive Optibond FL (Kerr). The restorations were evaluated after 6, 12, 24, and 36 months of clinical service regarding retention, marginal adaptation, microleakage, caries occurrence, and sensitivity. After a medium-long period of 3 yr, similar success in clinical performance was observed for the simplified all-in-one adhesive and the conventional three-step adhesive. However, the one-step adhesive exhibited significantly more incisal marginal defects and discolorations. Whereas marginal degradation appeared to arrest for the multistep etch-and-rinse adhesive after 12 months, the enamel margins of the restorations bonded with the one-step selfetch adhesive continued to deteriorate. These incisal marginal defects were, however, small and could easily be removed by polishing. For both adhesives, large and sclerosed lesions appeared to be at higher risk of retention loss.