Introduction Prospective studies reporting associations between cognitive performance and subsequent incident dementia have been subject to attrition bias. Furthermore, the extent to which established risk factors account for such associations requires further elucidation. Methods We used UK Biobank baseline cognitive data (n ≤ 488,130) and electronically linked hospital inpatient and death records during three- to eight-year follow-up, to estimate risk of total dementia (n = 1051), Alzheimer's disease (n = 352), and vascular dementia (n = 169) according to four brief cognitive tasks, with/without adjustment for constitutional and modifiable risk factors. Results We found associations of cognitive task performance with all-cause and cause-specific dementia (P < .01); these were not accounted for by established risk factors. Cognitive data added up to 5% to the discriminative accuracy of receiver operating characteristic curve models; areas under the curve ranged from 82% to 86%. Discussion This study offers robust evidence that brief cognitive testing could be a valuable addition to dementia prediction models.