The hospital cardiac arrest team is summoned in response to a sudden severe deterioration in a patient's condition. However, clinical experience suggests that some calls to general wards are preceded by a more gradual, possibly treatable decline. We undertook this study to define the extent of the problem and look for features that might enable prediction and prevention of cardiorespiratory arrest. We identified patients on general medical and surgical wards for whom cardiac arrest calls had been made. Their casenotes were examined for documentation of abnormal physical signs and laboratory test results in the 24 h before the call. We noted what doctors and nurses had done after abnormalities had been found. Over a 28-week period, calls were made for 47 patients on these wards. Twenty-four (51%) had premonitory signs. These patients were also less likely to survive to hospital discharge (P = 0.02). We conclude that some cardiorespiratory arrests are predictable. Wider appreciation of the significance of abnormal signs and laboratory test results could lead to prompter involvement of experienced clinicians and more aggressive therapy. Alternatively, as mortality is so high in this group of patients, more patients could be appropriately designated 'not for resuscitation'.