This study examined endosseous cylinder implant survival, defined as the unqualified presence of the implant in the mouth at the end of the observation period, in 598 consecutive VA patients, with a total of 2098 implants. Data were taken from the Department of Veterans Affairs (VA) Dental Implant Registry, which has maintained longitudinal data on the survival of individual dental implants in VA patients since 1987. The maximum time of observation in any one patient was 2040 days (5.6 yr). Survival analysis by use of life-table methods was carried out on both an implant- and a patient-specific basis. Implant cases were accrued randomly, and therefore a random censoring model was used. A correlated binomial model was used for assessment of the degree of within-patient clustering of implant removals. Results showed that the implant-specific survival rate during the longest time interval (5.6 yr) was 89.9%; the patient-specific implant survival rate during the same time was 78.2%. Among implants which were removed, the mean time to removal was 292 days. The hazard function, which describes the probability of implant loss as a function of time, decreased steadily throughout the observation period. The correlated binomial model suggested a clustering of removals within patients with multiple implants (rho = 0.11, p = 0.0001). The odds of having a second implant removed were 1.3 times greater if the patient had already had one implant removed. This study suggests that when implants fail, they do so soon after placement, and the likelihood of failure decreases steadily from implantation through the first five years post-surgery.(ABSTRACT TRUNCATED AT 250 WORDS)