The conceptual framework expects that caregivers who are fearful of becoming violent and who actually become violent toward elderly care recipients care for more seriously impaired elders; provide extensive help; care for relatives who exhibit more disruptive behaviors; experience aggression from the care recipient; have lower self-esteem; are younger; are more socially isolated; are the spouse of the recipient; and live with the care recipient. The data used to test these hypotheses were collected between January 1989 and October 1991 during face-to-face interviews with primary caregivers to elderly relatives who had been diagnosed with Alzheimer's disease or another nonreversible dementia within the previous 6 months. The dependent variable, violent feelings and actual violence was measured through carefully worded questions. The independent variables measured were caregiving demands, interactional stressors, caregiver characteristics, and caregiving context. The analyses confirmed several of the hypothesized risk factors. In the case of fear of violence, interactional stressors emerged as important predictors; both violence by the care recipient and disruptive behaviors increased the likelihood of violent feelings. Respondents with lower self-esteem were more likely to fear becoming violent. Living with the care recipients was also positively related to violent feelings. In differentiating between persons who only feared becoming violent and those who actually behaved violently, three factors emerged: interactional stress, being a spousal caregiver rather than another relative, and being an older caregiver. 3 tables and 52 references