The role of social factors in diabetes onset has been obscured by wide variation in their conceptualization and operationalization. We apply 3 theoretical frameworks to categorize social relationship variables along several dimensions and identify which dimension(s) are robustly associated with incident diabetes in the older adult population. The National Social Life, Health, and Aging Project (n = 2,365) and the Health and Retirement Study (n =11,824) provided longitudinal data from 57 to 90-year-old respondents over a 4- to 5-year period. Logistic regression models were used to test associations of 15 social variables measured identically in both data sets with diabetes onset measured as respondents' first report of a physician's diagnosis. In both studies, not being married, experiencing strain in a spousal relationship, and feeling lonely were associated with increased risk for diabetes onset at follow-up. Inconsistent or null findings were observed for social support, social activity, network size, number of friends and relatives, living alone, and closeness to network members. Robust findings in 2 large-scale surveys support the importance of the valence dimension (i.e., positive and negative); specifically, alleviating negative aspects of social life might more effectively reduce risk for diabetes than augmenting positive ones. Findings were not aligned with social variables differing on the subjectivity dimension (i.e., structural, functional, and qualitative aspects of social connections). Future work needs consistent conceptualization and measurement of social factors to correctly identify and categorize risk factors for diabetes onset and other health conditions in older adults.