Answers to standardised attitude questions not only depend on the attitude respondents have on the issue, but are also influenced by the wording of the question. A well known example of this phenomenon is the forbid/allow asymmetry. Although the verbs 'forbid' and 'allow' are supposed to be each others' counterparts, the answers to questions worded with the verb 'forbid' turn out not to be opposite to answers to equivalent 'allow'-questions: respondents turn out to be more likely to respond 'no' rather than 'yes' to 'forbid' questions. It is commonly assumed that this is caused by the extreme connotations of both verbs. In this article a meta-analysis is reported over all forbid/allow research since 1940. First of all it is analysed whether the asymmetry can be generalised over questions and experiments. This turns out to be the case. The answer 'no, not forbid' is obtained more than 'yes, allow', and the mean effect size turns out to be quite large. The huge variance in asymmetry size however, offers room for a search after additional explanations. It turns out that the interaction between the complexity of the issue in the question, and the degree of ness of the question text, is systematically related to the size of the asymmetry. This stresses the importance of attitude strength as an explanatory variable, as well as the importance of looking at the communicative task as a whole. © Swets & Zeitlinger.