This article investigates possible sources of error in a public opinion survey conducted by a metropolitan police service. Logistic regression models are used to analyse the results of a mixed-mode survey in which respondents were questioned over the telephone or through a Web survey. Even after controlling for demographics, respondent lifestyle and criminal victimization, Web respondents were significantly more likely to (i) give lower scores on questions about satisfaction with the police and level of perceived police legitimacy, (ii) indicate unfavourable opinions about the police, (iii) confess worries or fear about their own security and (iv) select moderate response choices. Telephone surveys seem to overestimate favourable opinions about the police and underestimate public concerns about security. It is important to understand these effects because survey results are often used to support strategic decisions about public safety or to assess police performance.