Purpose: To compare the 3-year outcomes of immediately loaded and one-stage conventionally loaded variable-thread tapered implants in the posterior maxilla. Materials and methods: This study was designed as a split-mouth randomised controlled trial. Twenty-six patients attending the postgraduate Periodontics Department at the Lebanese University, and missing teeth bilaterally in the posterior maxilla were randomised. All patients received three to four implants in each of the posterior sextants. The implants on one side were immediately loaded with a provisional resin fixed partial denture on definitive multi-unit abutments regardless of their primary stability. The implants in the contralateral side received definitive multi-unit abutments according to the one-stage unloaded protocol. Three to 3.5 months following implant placement, the implants were restored with metal-ceramic fixed prostheses. Outcome measures were implant and prosthesis failure rates, complications, and peri-implant bone level changes up to 3 years following delivery of the definitive prosthesis. The clinical outcomes and radiographic measurements were performed by a single outcome assessor blinded to the type of interventions. Results: Two patients dropped out prior to the delivery of definitive prostheses. The results were analysed using a per-protocol analysis and included 24 patients. Four implants supporting a four-unit immediately loaded prosthesis failed in one patient, 3 months following definitive prosthesis. In the same patient, the three contralateral conventionally loaded implants failed 14 months after definitive prosthesis. There were no significant differences in the proportions of implant and prosthesis failures at 3 years (difference = 0%; 95% CI 0.0% to 14.2%; P = 0.999). Peri-implantitis was diagnosed at two adjacent conventionally loaded implants in one patient at the 3-year examination. In the immediately loaded group, four early minor prosthetic complications occurred during the provisionalisation phase. Following delivery of the definitive prostheses, one minor ceramic fracture was observed in each of the implant groups. The difference in the rate of complications between the two interventions was not statistically significant at 3 years (difference = 13%; 95% CI 3.4% to 27.7%; P = 0.453). The 3-year peri-implant marginal bone level changes were evaluated in 23 patients (77 immediately loaded and 76 conventionally loaded implants). On average, patients lost 0.79 (0.62) mm at the immediately loaded and 0.91 (0.82) mm at the conventionally loaded implants, the difference being statistically not significant (difference = 0.12 mm; 95% CI −0.31 to 0.55 mm; P = 0.590). The 3-year marginal bone level changes were not significantly different between smokers (n = 12) and non-smokers (n = 11) (difference = 0.19 mm; 95% CI −0.24 to 0.62 mm; P = 0.382). Conclusions: Immediate loading of three- to four-unit fixed partial prostheses supported by variable-thread implants in the posterior maxilla can achieve similar 3-year results to one-stage conventionally loaded implants.