PurposeTo investigate the behaviour patterns of typical noise sources in critical care wards, and relate their patterns to healthcare environment in which the sources adapt themselves in several different forms. Methods An effective observation approach was designed for noise behaviour in the critical care environment. Five descriptors have been identified for the behaviour observations, namely interval, frequency, duration, perceived loudness and location. Both the single-bed and multiple-bed wards at the selected Critical Care Department were randomly observed for three inconsecutive nights, from 11:30PM to 7:00AM the following morning. The Matlab distribution fitting tool was applied afterwards to plot several types of distributions, and estimate the corresponding parameters. Results The lognormal distribution was considered to be the most appropriate statistical distribution for noise behaviours in terms of the interval and duration patterns. The turning of patients by staff was closely related to the increasing occurrences of noises. Among the observed noises, talking was identified with the highest frequency, shortest intervals and the longest durations, followed by monitor alarms. The perceived loudness of talking in the night time wards was classified into 3 levels (raised, normal and low). The majority of people engaged in verbal communication in the single-bed wards occurred around the Entrance Zone, whereas talking in the multiple-bed wards was more likely to be situated in the Staff Work Zone. As expected, more occurrences of noises along with longer duration were observed in multiple-bed wards rather than single-bed wards. 'Monitor plus ventilator alarms' was the most commonly observed combination of multiple noises.