To investigate the effect of stringent and lenient criteria upon the process of item selection in the development of a health status questionnaire, an item pool (179 items) was administered to 139 patients with left ventricular dysfunction. Associations between each item and the criteria of gender, age, duration of disease, global health and global impairment were examined. Items were selected from the pool on the basis of their associations with the criteria using four levels of stringency. The most stringent criteria rejected items which had a shared variance of > or = 4% with gender, age and duration of disease and a shared variance of > or = 6% with global health and impairment. The most lenient criteria rejected items which had a shared variance of > or = 6% with gender, age and duration of disease and a shared variance of > or = 4% with global health and impairment. Using the most stringent criteria, 75 items were selected, compared with 127 items using the most lenient criteria. Small differences in the level of association had large effects on item selection. The choice of level of association used to base item selection can have a crucial influence on questionnaire content.