Purpose: To evaluate the diagnostic accuracy of bone loss/bone levels, as detected after 5 years of implant function, in identifying patients who suffered implant failures in the following 5 years. Materials and Methods: Data on radiographic measurements of marginal bone levels at prosthesis placement and after 5 years of function were retrospectively retrieved from 11 previous publications. Included patients were allocated into different subgroups with regard to bone loss/bone level during/after 5 years in function, respectively. A diagnostic test was used to estimate the accuracy of finding patients/jaws/implants at risk for a future implant failure by calculating sensitivity, specificity, positive predictive values (PPVs), and negative predictive values (NPVs) in different subgroups, respectively. Results: Altogether, 749 treated jaws (723 patients/3,363 implants) were included in the study group. Treated jaws in the high-level subgroups presented an overall higher risk of implant failures from 5 to 10 years in function (P < .05). Many treated jaws/implants were allocated into the high-level groups, but the proportions of implant failures were low in these groups. The diagnostic test comparing high- and low-level groups with and without implant failures showed low accuracy to predict implant failures; the PPV ranged from 4% to 33%. Lower PPVs were observed for diagnostic tests for individual implants (range: 4% to 6%). Conclusion: More severe bone loss was associated with higher risk of future implant failure. However, many patients/ implants with obvious bone loss in the study group and low prevalence of implant failures at the 10-year examination resulted in poor accuracy in identifying individual patients or implants at risk for failure. This suggests that it is difficult to predict future implant failures based only on radiographic measurements.