Questionnaires that are used in studies with severely patients should be as short as possible. Abridged versions of existing inventories are very practical in these instances. The answers of 444 subjects in three samples (cancer patients, medical students, surgical patients) were used to investigate the possibility of constructing short and reliable versions of the scales of the State and Trait Anxiety Inventory. A stepwise regression procedure showed the possibility to reliably predict the total score of the unabridged versions by means of weighted sums of eight items for each scale. Omission of weights did not lead to substantial loss of information. Cronbach's alpha of the State-scale decreased from 0.93 to about 0.85 for different combinations of items and from 0.91 to about 0.82 for combinations of eight items of the Trait-scale. The relationship between both scales was only slightly modified by the shortening procedure.