We have tested the hypothesis in this study that different methods of administering a questionnaire produce differential approximations to truthful admission of undesirable personality traits and behaviours. Four different methods of administration produced different levels of trust and understanding, using the current prediction among healthy subjects of death by cancer or coronary heart disease 13 years later as the criterion. There were significant differences in the accuracy of the predictions, depending crucially on the method of administration of the questionnaires. Best predictions were achieved for subjects when both trust and understanding had been increased by interviewers' suitable participation; worst results were achieved for subjects when no special effort was made to increase either. Intermediate results were found for procedures which increased either trust or understanding. It is argued that the success or failure of studies investigating the influence of personality and stress on diseases like cancer and coronary heart disease may depend crucially on the adopted method of interrogation.