Purpose: The aim of this study was to compare the accuracy of digital dental impressions with the accuracy of impressions obtained via conventional techniques. Methods: Two different master models were created, one with parallel implants (model 1) and the other with non-parallel implants (model 2). These reference master models included 4 Klockner KL RP implants (Klockner Implant System SA, Barcelona, Spain), which were juxta-placed and equidistant in the intermentoneal region. In model 1 the implants were placed parallel to each other, whereas in model 2 the implants were placed such that there was a divergence angle of 15° between the more distal implants, and a convergence angle of 15° between the two central implants. A total of four types of impressions were obtained from model 1 (four groups, n = 10 each), including closed tray impressions with replacement abutments; open tray impression groups for dragging copings, without splinting; open tray impressions for ferrules; and impressions obtained using the 3MTM True Definition Scanner system. For model 2 three groups were created (three groups, n = 10 each), including closed tray impressions with replacement abutments; open tray impression for dragging copings, without splinting; and impressions obtained using the 3MTM True Definition Scanner system. The master models and the models obtained using conventional methods were digitalized in order to compare them via an extraoral high-resolution scanner (Imetric IScan D104i, Porretruy, Switzerland). The STL (Stereo Lithography (format for transferring 3 dimensional shape information)) digital values were loaded into reverse-engineering software and superimposed with their respective STL master models in order to evaluate deviations in three dimensions. We then analyzed the squares of the deviations in the three axes and evaluated the median and the sum of the deviation square. Statistical analysis was performed using the IBM Corp. Released 2016. IBM SPSS Statistics for Windows, Version 24.0. Armonk, NY: IBM Corp. The normality of the distributions was analyzed according to a Kolmogorov-Smirnov test. The median comparison was performed using the differences between the medians, analyzed using non-parametric Kruskal-Wallis and Mann-Whitney tests with a significance level of p < 0.05. Results: For model 1, the deviations of the digital impressions were smaller than those associated with the conventional techniques. The sum value in group D was 1,068,292, which was significantly lower than those of groups A, B, and C, which were shown to be 2,114,342, 2,165,491, and 1,265,918, respectively. This improvement was not observed when using model 2, however, where the conventional techniques yielded similar results. Group F simultaneously presented the lowest total square sum of the three deviations (1,257,835), indicating a significantly higher accuracy for this group in model 2, while the sum values were 1,660,975 and 1,489,328 for groups E and G, respectively. Conclusion: Digital impressions of full-arch models were abl...