Previous investigations of recall and recognition for threatening information in clinically anxious subjects have yielded equivocal results. The present study contrasts implicit (word completion) with explicit (cued recall) memory and shows that indices of bias for emotional material derived from the two types of memory are independent of one another. The explicit measure was correlated with trait anxiety scores, but did not clearly distinguish between subjects with clinical anxiety states and normal control subjects. On the implicit memory measure, clinically anxious subjects produced more threat word completions, but only from a set to which they had recently been exposed. These results are taken as evidence that internal representations of threat words are more readily or more persistently activated in anxiety states, although they are not necessarily better elaborated.