The purpose of this study was to assess in vitro the marginal fit of four-unit fixed partial dentures (FPDs) produced using three different computer aided design/computer aided manufacturing (CAD/CAM) all-ceramic systems before and after porcelain firing cycles and after glaze cycles. An acrylic resin model of a maxillary arch was fabricated. Teeth #6 and 9 were prepared; teeth #7 and 8 were absent. Forty-five four-unit zirconium-oxide-based ceramic FPDs were made following conventional impression and master cast techniques: 15 were made with the Everest system, 15 with the Procera system, and 15 with the Lava system. Marginal gaps along vertical planes were measured for each bridge before (Time 0) and after (Time 1) porcelain firing cycles and after glaze cycles (Time 2) using a total of 8 landmarks (4 for tooth #6 and 4 for tooth #9) by means of a microscope at a magnification of x50. MANOVA was performed to determine whether the 8 landmarks, jointly considered, differed between CAD/CAM systems and time phases. Two-way ANOVA was performed to investigate in detail, for each landmark, how gap measurements were related to CAD/CAM systems and time phases. Differences were considered to be significant at p < 0.05. The mean values of the Everest system (microm) were: 63.37 (Time 0), 65.34 (Time 1), and 65.49 (Time 2); the mean values of the Lava system (microm) were: 46.30 (Time 0), 46.79 (Time 1), and 47.28 (Time 2); the mean values of the Procera system (microm) were: 61.08 (Time 0), 62.46 (Time 1), and 63.46 (Time 2). MANOVA revealed quantitative differences of the 8 landmarks, jointly considered, between the three CAD/CAM systems (p < 0.0001), but it did not reveal any quantitative differences among the three time phases (p > 0.4). Two-way ANOVA revealed that the Lava system produced gap measurements statistically smaller than the Everest and Procera systems (p < 0.0001 for each landmark). Within the limitations of this study, it was concluded that the three zirconium-oxide-based ceramic CAD/CAM systems demonstrated a comparable and acceptable marginal fit; however, the Lava system produced gap measurements statistically smaller than the Everest and Procera systems. The porcelain firing cycles and the glaze cycles did not affect the marginal fit of the zirconium-oxide-based ceramic CAD/CAM systems.