This paper examines the methods used in expressing agreement between observers both when individual occurrences and total frequencies of behaviour are considered. It discusses correlational methods of deriving inter-observer reliability and then examines the relations between these three methods. Some of the factors that affect reliability are reported. These include problems of definition such as how a behaviour may change with age and how reliability depends on which recipient of behaviour is considered. Frequency of occurrence of a behaviour pattern is discussed both in relation to itself and to other behavioural categories, as well as its effect on observer vigilance. The effect of partitioning data in different ways is also discussed. Finally the influence of different observers on inter-observer reliability is noted. Examples are taken from studies of children and kittens.