The possibilities of computer-based cognitive training (CCT) in postponing the onset of dementia are currently unclear, but promising. Our aim is to investigate older adults´ adherence to a long-term CCT program, and which participant characteristics are associated with adherence to the CCT. This study was part of the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER). Participants were 60-77-year-old individuals with increased dementia risk, recruited from previous population-based studies. The participants included in this study (n = 631) had been randomized to receive a multi-domain lifestyle intervention, including CCT. The measure of adherence was the number of completed CCT sessions (max = 144) as continuous measure. Due to a substantial proportion of participants with 0 sessions, the zero inflated negative binomial regression analyses were used to enable assessment of both predictors of starting the training and predictors of completing a higher number of training sessions. Several cognitive, demographic, lifestyle, and health-related variables were examined as potential predictors of adherence to CCT. Altogether, 63% of the participants participated in the CCT at least once, 20% completed at least half of the training, and 12% completed all sessions. Previous experience with computers, being married or cohabiting, better memory performance, and positive expectations toward the study predicted greater odds for starting CCT. Previous computer use was the only factor associated with a greater number of training sessions completed. Our study shows that there is a large variation in adherence to a long-lasting CCT among older adults with an increased risk of dementia. The results indicate that encouraging computer use, and taking into account the level of cognitive functioning, may help boost adherence to CCT.