Self-rated health, the most widespread measure of health in questionnaires, is merely a biased reflection of real health. However, it is difficult to isolate and assess this bias. In particular, to what extent does the information collected depended on the collection conditions, given that these vary considerably between countries and surveys? The resources in the Share database enable us to analyse two collection method effects: the position of the rated health question in the questionnaire and the wording of the suggested response modalities. The two effects have an impact: the average level of health, the distribution of states of health and links between health and other variables are altered when the collection conditions are changed. The positioning effect is homogenous: in all countries in the sample, individuals on average give higher levels of rated health after answering a detailed questionnaire focusing on their real state of health compared with the levels indicated when the question is asked beforehand. Respondents also seem to give greater importance to the functional dimension of health ­ mobility and everyday, social and sensorial restrictions ­ when the rated health variable is placed after the detailed questionnaire. The wording of the suggested response modalities also changes the measure of self-rated health. For example, Sweden appears to be the healthiest country (out of ten countries) using an initial wording, but moves down to fifth place using a second. Moreover, we have shown that the wording chosen by the Rand Corporation for its Short Form 36 questionnaire, which has since been used in most questionnaires in English-speaking countries, was no more discriminating, in terms of real health, than the wording widely used in health questionnaires in continental Europe.