Statement of problem Dental implant geometry affects the mechanical performance and fatigue behavior of butt-joint implant-supported restorations. However, failure of the implant component has been generally studied by ignoring the prosthetic screw, which is frequently the critical restoration component. Purpose The purpose of this in vitro study was to evaluate the effect of 3 main implant geometric parameters: the implant body diameter, the platform diameter, and the implant-abutment connection type (external versus internal butt-joint) on the fatigue life of the prosthetic screw. The experimental values were further compared with the theoretical ones obtained by using a previously published methodology. Material and methods Four different designs of direct-to-implant dental restorations from the manufacturer BTI were tested. Forty-eight fatigue tests were performed in an axial fatigue testing machine according to the International Organization for Standardization (ISO) 14801. Linear regression models, 95% interval confidence bands for the linear regression, and 95% prediction intervals of the fatigue load-life (F-N) results were obtained and compared through an analysis of covariance (ANCOVA) to determine the influence of the 3 parameters under study on the fatigue behavior (α=.05). Results Linear regression models showed a statistical difference (P<.001) when the implant body diameter was increased by 1 mm; an average 3.5-fold increase in fatigue life was observed. Increasing the implant abutment connection diameter by 1.4 mm also showed a significant difference (P<.001), leading to 7-fold longer fatigue life on average. No significant statistical evidence was found to demonstrate a difference in fatigue life between internal and external implant-abutment connection types. Conclusions Increasing the implant platform and body diameter significantly improved (P<.001) the fatigue life of the prosthetic screw, whereas external and internal connections provided similar results. In addition, experimental results proved the accuracy of the fatigue life prediction methodology.< Réduire