Few data are available on the effect of social ties on dementia development. This study explored whether single social network components and different degrees of the social connections affect dementia incidence. A community-based cohort of 1203 non-demented people, living at home in the Kungsholmen district of Stockholm, Sweden, and who had good cognition, was followed for an average period of 3 years. On the basis of medical and psychological data, 176 patients were diagnosed with dementia according to the criteria of the third edition revision of the Diagnostic and Statistical Manual of Mental Disorders. Information on social network was obtained by personal interview by trained nurses at baseline. The covariates included in the analysis were age, sex, education, cognitive and functional status, depressive symptoms, and vascular diseases. Those individuals living alone, and those without any close social ties, both had an adjusted relative risk for developing dementia of 1.5 (95% CI 1.0-2.1 and 1.0-2.4, respectively). Compared with married people living with someone, single people and those living alone had an adjusted relative risk of 1.9 (95% CI 1.2-3.1). Infrequent contacts with network resources did not increase the risk of the disease if such contacts were experienced as satisfying. When all components were combined in an index, a poor or limited social network increased the risk of dementia by 60% (95% CI 1.2-2.1), and a significant gradient was found for the four degrees of social connections (p=0.0009). An extensive social network seems to protect against dementia. Confirmation of this finding and further investigation to clarify the mechanisms are worthwhile due to the implications for prevention.