The aim of this study was to determine if the order of questions in a study on men who have lost a wife in cancer affects self-assessed measures of psychological morbidity. Data were collected from 76 men who had lost a wife owing to cancer in the breast, ovary or colon in 1999. They were randomly allocated to one of two questionnaires with identical content but varying design. One version began with questions about the wife's disease and ended with the respondent's current wellbeing (morbidity-last group). The other version had a reversed order (morbidity-first group). Results showed that the design of the questionnaire affected self-assessed psychological morbidity; all relative risks for these measures were above 1.0 in the morbidity-last group. The highest relative risk was obtained for anxiety (as measured by HADS), 3.4 (0.8-15.0), and depression (as measured by a visual-digital scale), 3.1 (1.2-8.5). Psychological morbidity is assessed as higher when questions appear in the end, rather than the beginning, of a bereavement-related questionnaire. In order to avoid a detrimental bias in a study on bereavement, psychological morbidity is probably best measured first in such a questionnaire.