Alarms are used in many clinical applications, but they are often less than optimal because the design and implementation of alarms has not always taken the cognitive capacity and processing mechanisms of the user into account. As a result alarms are frequently too loud, irritating, confusing, badly designed, and too numerous, resulting in them often being turned off and hindering, rather than enhancing, task performance. This paper reviews some of the main areas where it is essential to take account of the cognitive system of the user and behavioural processes more generally. Five central areas of concern are discussed: the number of alarms and ways that this might be reduced; false alarm rates and their impact on human responses; the design of alarms and the application of research into auditory cognition on design; intelligent alarm systems; and the proposals for alarm design set out in a recent worldwide medical alarms standard. In each area some background is given and the implications for alarm design and implementation outlined. The conclusion is that there are some indications that alarm design and implementation takes account of relevant research data, but that there is still some way to go before these findings are fully integrated and the situation is improved upon further.