An item analysis was carried out on Forms A (n = 200) and B (n = 252) of the Eysenck Personality Inventory (EPI) in order to establish how well the items perform. The method was, for each scale, to compare the mean total score of the people who said ‘yes’ to each item with the mean total score for the people who said ‘no’. The differences between the means were tested for significance, and it was found that some items had no significant differences, some differed significantly on the expected scale, some on the expected scale as well as others, some better on scales other than the expected one, and some did not differ on the expected scale but did on another scale. Particularly useful was the discovery of items which when answered ‘no’ predict the neuroticism score, since such items are not included in the EPI. On the basis of the analysis, various tests could be constructed. Useful tests might contain a better balance of ‘yes’ and ‘no’ items, items which predict the total scores better than does the average item, or only the very best items to make up a very short form for the selection of extreme groups.