STUDY OBJECTIVES--To estimate the population based requirement for neonatal intensive care (NIC) cots by investigating NIC utilisation in a large population based study. DESIGN AND SETTING--This was a two year, non-randomised, prospective cohort study of neonates receiving NIC in hospitals in the Trent Regional Health Authority (RHA). PARTICIPANTS--The main study considered the 2979 neonates born to Trent RHA residents who had begun high dependency care in Trent RHA or neighbouring health authorities between 1 April 1990 and 31 March 1992 and met any of the following criteria: (i) birth weight < 1500 g; (ii) gestation < 32 weeks; (iii) need for active respiratory support other than initial resuscitation; (iv) need for in utero or neonatal transfer to receive high dependency care; (v) severe asphyxial brain insult after delivery : and (vi) death. The analysis here is restricted to the 1730 neonates who received total parenteral nutrition or assisted ventilation, or both; that is, those who received NIC level 1 (ICL1) on at least one day. MEASUREMENT AND ANALYSIS--The treatment history of each neonate was converted into a time-line detailing the dates of beginning and stopping NIC, the dates of any transfers between units, and any gaps in NIC treatment. The duration of ICL1 was observed directly and that of intensive care level 2 (ICL20 was imputed from a model based upon round trip transfers. These were also recorded on the timeline. The time-lines were first used to derived the observed distribution of the utilisation of NIC cots in Trent RHA during the study. An extensive series of Monte Carlo simulations was then carried out in order; (i) to estimate cot requirements in smaller populations; (ii) to determine whether Trent RHA utilised its NIC cots in a manner that was clinically appropriate at the population based level; (iii) to perform a series of sensitivity analyses; and (iv) to compare results with an equivalent study carried out in the Northern RHA. MAIN RESULTS AND CONCLUSIONS--Trent RHA is reasonably representative of the greater population of England and Wales in terms of both the distribution of birth weight and of birth weight-specific neonatal mortality. Trent RHA did not seen to be underprovided for NIC cots or to be overusing these cots inappropriately. It therefore seems reasonable, if the assumptions of the analysis are borne carefully in mind, to treat these utilisation data as a rough guide to true population based need. NIC cot requirements depend critically upon the size of the served population - small populations are subject to greater random variability and require relatively more cots to ensure cot availability on an equivalent proportion of days. A neonatal unit should not therefore serve a population generating fewer than 5000 and 25 000 births per annum, the estimated population based provision which would ensure free cots on 29 out of 30 days falls gradually from 1.20 to 0.88 NIC cots per 1000 births per annum. A cooperative network of NICUs offers the opportunity to provide fewer cots per head of population while maintaining good access for most neonates referred to the service.