The purpose of this thesis is to examine risk factors present before the diagnosis of dementia in the Aging, Demographics and Memory Study (ADAMS), a sample of 856 participants chosen from the Health and Retirement Study (HRS), a nationally representative sample of persons of retiring age and older, to take part in a clinical assessment for cognitive impairment and collection of other health information. A subset of 330 individuals from the ADAMS study diagnosed with either Alzheimer¡¦s disease (AD) or vascular dementia (VaD) was considered in the primary analyses. Risk factors examined were age, gender, years of education, APO-fÕ status, and empirical Bayes estimates of latent growth curve trajectory components of longitudinal episodic memory performance, mental status and cardiovascular risk to test whether it was possible to discriminate whether a participant would be later diagnosed with either Alzheimer¡¦s disease (AD) or vascular dementia (VaD). Data from the HRS from up to a decade before diagnosis were used in logistic regression analyses to find the best fitting model of prediction into groups of either AD or VaD. Results showed that while age, gender, number of APO-fÕf¤alleles, episodic memory and cardiovascular risk factors were predictive of later diagnosis of AD versus VaD subtypes, educational attainment and longitudinal mental status trajectories were not significant predictors. Each APO-fÕf¤ allele more than doubled the odds of being classified into the AD group (OR =2.48). Higher levels of performance and maintenance of episodic memory ability across age decreased the odds of being classified in the AD group (ORIntercept = 0.92; ORSlope= 0.79). Every unit of increased cardiovascular risk tended to decrease the odds of being classified into the AD group (OR = 0.77). An attempt was made to examine mixed dementia cases by a re-categorization of participants with vascular pathology into new groups of mixed cases versus a more ¡¥pure¡¦ AD group but the percent of cases that were correctly classified decreased from 79.7% in the original analyses to 77.9% once re-organized, indicating more may need to be done to get at underlying risk and cognitive factors involved in mixed dementia.