Between 1997 and 2001, 17 teaching and 56 non-teaching acute English hospitals conducted hospital-wide surveillance of hospital-acquired bacteraemia (HAB) using a standard protocol drawn up by the Nosocomial Infection National Surveillance Scheme (NINSS). The sources of organisms, the incidence of device-related HAB, and the distribution of HABs from individual device-related sources by specialty and type of hospital were determined for 6,956 HABs in order to identify where resources should best be targeted to reduce these infections. The overall incidence of HAB was higher in teaching than in non-teaching hospitals: 5.39 and 2.83 HABs per 1,000 patients at risk, respectively (P<0.001). Device-related sources were responsible for 52.4 and 43.2% of all HABs in teaching and non-teaching hospitals, respectively (P<0.001), and central lines were the commonest source, causing 38.3% of HABs in teaching versus 22.3% in non-teaching hospitals (P<0.001). In teaching hospitals, general intensive care units (ICUs), haematology, special care baby units (SCBUs), nephrology, and oncology accounted for only 6.1% of the population surveyed, but had the highest incidence of HAB, and contributed 47.8% of 2091 HABs and 56.9% of 1,095 device-related bacteraemias. Of 623 device-related bacteraemias in these high-risk specialties, 554 (88.9%) were from central lines. Thus, in teaching hospitals, resources should be targeted primarily at the prevention of central line-related bacteraemia in these five high-risk specialties, and the surveillance should include data on central line use. In non-teaching hospitals, nearly two thirds (63.3%) of 4,865 HABs and 60.7% of 2,103 device-related bacteraemias were from a few specialties with a low incidence of bacteraemia, but large numbers of patients, namely general medicine, general surgery, geriatric medicine and urology. These specialties accounted for 50.5% of the population surveyed. Central lines were the most common source of bacteraemia in general medicine and surgery, and together accounted for 23.3% of all device-related bacteraemias. However, in geriatric medicine and urology, central line sources were infrequent, accounting for only 1.7% of all device-related bacteraemias. On the other hand, bacteraemia from catheter-associated UTI were common in all these four specialties accounting for 20.9% of all device-related bacteraemias. Thus, in non-teaching hospitals, resources should be targeted primarily at these low-risk specialties and surveillance should include, at least, bacteraemia from central lines and from catheter-associated UTI. Further benefit can be obtained by including central line-related bacteraemias from general ICU and haematology patients, as they contributed 17.0% of all device-related bacteraemias in non-teaching hospitals.