The aim of this study was to evaluate the change of marginal bone level radiographically around three different implant systems after 3 years in function. Fifty-four patients were included and randomly assigned to three treatment groups of rough-surface implants (TiUnite, n = 37), hybrid of smooth and rough-surface implants (Restore, n = 38) and rough surface with microthread implants (Hexplant, n = 45). Clinical and radiographic examinations were conducted at the time of implant loading (baseline), 1 and 3 years after loading. A three-level mixed-effect analysis of covariance (ancova) was used to test the significance of the mean marginal bone change of the three implant groups. A total 120 of 135 implants completed the study. None of the implants failed to integrate. Significant differences were noted in the marginal bone loss recorded for the three groups (P < 0.0001). At 3 years, the rough surface with microthread implants had a mean crestal bone loss of 0.59 +/- 0.30 mm; the rough-surface implants, 0.95 +/- 0.27 mm; and the hybrid surface implants, 1.05 +/- 0.34 mm. Within the limitations of this study, rough-surface implants with microthread at the coronal part might have a long-term positive effect in maintaining the marginal bone level against functional loading in comparison with implants without these two features.