The accuracy with which testees perceived the difficulty of ability-test items was investigated by administering two 41-item conventional tests on verbal ability. High- and low-ability groups responded to test items by choosing the correct alternative and then rating each item's difficulty relative to their levels of ability. Least-squares estimates of item difficulty correlated highly with proportion-correct and latent trait estimates of item difficulty based on a norming sample. Least-squares estimates of testee ability, which were based solely on the difficulty perceptions of the testees, correlated significantly with number-correct and maximum-likelihood ability scores based on the testees' responses to the items. These results show that item-difficulty often used in test construction, and that as testee ability level increased, the items were perceived as being relatively less difficult. The relationship between a testee's ability and his/her perception of an individual item's relative difficulty appeared to be weak. Of major importance was the finding that items which were appropriate in difficulty levels from a psychometric standpoint were perceived by the testees as being too difficult for their ability levels. (Author/DSE)