Purpose: This study evaluated the accuracy of 2 implant-level impression techniques (direct nonsplinted and splinted) for the fabrication of multi-unit internal-connection implant restorations in 2 simulated clinical settings (parallel and divergent) using a laboratory model. Materials and Methods: A dental stone master model was fabricated with 2 pairs of implant replicas. One pair simulated a parallel clinical condition and the other an 8-degree-divergent condition. Ten stone casts were made from vinyl polysiloxane impressions of the master model for each impression technique. Half of the samples were created by a direct nonsplinted technique (square impression copings, custom tray), and the other half were made by a direct splinted technique (square impression copings splinted with autopolymerizing acrylic resin, custom tray). Four strain gauges were fixed on each metal framework to measure the degree of framework deformation for each stone cast in half-Wheatstone-bridge formations. Deformation readings were made twice in 4 directions (anterior, posterior, superior, and inferior). Deformation data were analyzed using repeated-measures analysis of variance at a .05 level of significance. Results: No significant difference in deformation was found between the direct nonsplinted and splinted samples in either simulated clinical condition (P > .05). No significant difference in deformation was found between the techniques regardless of condition (P > .05). Conclusions: Within the limitations of this study, using a 2-implant model, the accuracy of implant-level impressions for internal-connection implant restorations was similar for the direct nonsplinted and splinted techniques in settings with divergence up to 8 degrees. Schlagwörter: implant divergence, implant-level impression, internal connection implant, strain gauge