(1) Background: The objective of this study was to assess which digitization method produces the biggest deviation in the 3D images of tooth size from plaster models made using alginate impressions, which are considered the gold standard in orthodontics. (2) Methods: The sample used in this study included 30 subjects (10 males and 20 females). Measurements were made on four types of models: (1) digital models obtained through intraoral scanning and digitized models of plaster cast made from (2) alginate impressions, (3) silicone impressions, and (4) conventional plaster models. Mesio-distal (MD) and buccal/labial–lingual/palatal (BL) dimensions were measured on the reference teeth of the right side of the jaw (central incisor, canine, first premolar, and first molar). Comparisons of tooth size between the methods were conducted using a repeated measurement analysis of variance and the Friedman test, while the intraclass correlation coefficient was used to determine agreement between the different methods. (3) Results: The results showed a similar level of agreement between the conventional and digital models in both jaws and the anterior, middle, and posterior segments. Better agreement was found for the MD measurements (r = 0.337–0.798; p ≤ 0.05) compared to the BL measurements (r = 0.016–0.542), with a smaller mean difference for MD (0.001–0.50 mm) compared to BL (0.02–1.48 mm) and a smaller measurement error for MD (0.20–0.39) compared to BL (0.38–0.89). There was more frequently a better level of agreement between 3D images than measurements made using a digital caliper on the plaster models with 3D images. (4) Conclusions: The differences in measurements between the digital models and conventional plaster models were small and clinically acceptable.