Despite the aim to count everyone, and the fact that there is a legal requirement on all individuals to complete a census form, censuses are always subject to some counting error. This is mainly a problem of under counting rather than over counting, although both are possible.
Undercounting: Enumerators may not identify all residential addresses where people are living, or they may fail to identify all households living at an address. Even where households receive forms, this does not guarantee an accurate count of household members. Those completing census forms may, deliberately or otherwise, miss off people normally resident at the address. Some may simply fail to return completed forms.
The 1991 census is believed to have missed around 4% of the population at the data collection phase. The problem of under-enumeration is at the heart of some important changes that were introduced for the 2001 Census (see What steps were taken to improve coverage in the 2001 Census?).
Over counting: Some over counting may occur where people living at more than one address are counted at both addresses. The process of adjusting census counts for under-enumeration may also inflate numbers where enumerators wrongly identify an 'absent' household where non exists
Despite these problems the census records a much better response rate than most other major social surveys. For example, the response rate for the 2000 General Household Survey was 68%.