The difficulty level choices made by examinees during a self-adapted test w
ere studied. A positive correlation between estimate ability and difficulty
choice was found. The mean difficulty level selected by the examinees incr
eased nonlinearly as the testing session progressed. Regression analyses sh
owed that the best predictors of difficulty choice were examinee ability, d
ifficulty of the previous item, and score on the previous item. Four strate
gies for selecting difficulty levels were examined, and examinees were clas
sified into subgroups based on the best-fitting strategy. The subgroups dif
fered with regard to ability, pretest anxiety, number of items passed, and
mean difficulty level chosen. The self-adapted test was found to reduce sta
te anxiety for only some of the strategy groups.