To measure the speed of response to ventricular fibrillation on general medical wards and to assess the importance of this and other factors for survival to leave hospital, 69 consecutive patients with ventricular fibrillation were studied prospectively using an automatic timing device in the hospital telephone exchange and an automatic timer and ECG recording during resuscitation. Twenty-seven patients were initially resuscitated and 17 were discharged from hospital. The median time to connect the monitor after recognition of a cardiac arrest was 127 s (range 0-277) for survivors and 132.5 s (range 0-620) for non-survivors. The median time ventricular fibrillation was displayed before the first shock was 43 s (range 4-75) for survivors and 52 s (range 10-454) for non-survivors. These differences were not significant; but logistic regression analysis identified primary ventricular fibrillation, short display time (logged data), 'early' time of day, absence of pre-existing non-cardiac illness, and post-defibrillation heart-rate > 30 beats.min-1 in rank order as independent predictors of survival. In spite of no significant diurnal variation in response time, successful resuscitations were concentrated in the early nursing shift (0730-1530 h). Four shocks were inappropriate. Clinical diagnosis was more predictive of outcome than the time to the first shock. The reasons for the poorer results in the evening and night are uncertain.