Currently, the limited options for sampling frames and sampling methods in the UK severely constrain the use of web in probability-based general population surveys. Good population coverage can be achieved through use of sampling frames  traditionally used for interviewer-administered surveys, but unless the initial approach is then interviewer-administered, the researcher loses control over respondent selection. Web data collection is mainly considered only for longitudinal surveys, in which respondent names are known and email addresses can be collected at previous waves. For one-time surveys a mail invitation to a web survey can be used but errors of respondent identification are likely and are hard to detect. The prospects for web data collection from the general population would change dramatically if a population register were to become available and especially if this were to include email addresses. However, this seems highly unlikely, at least in the foreseeable future. For these reasons, non-probability methods are likely to remain attractive to many for some time to come.