Two experiments were carried out to study the metamemory theory of confidence for the domain of sentence recall. Experiment 1 used nondeceptive sentences and deceptive synonym substitution sentences. Experiment 2 used nondeceptive sentences and deceptive schema inference sentences. In both experiments there was a strong positive relationship between memory confidence and memory accuracy for nondeceptive items and a much reduced relationship for deceptive items. The results were interpreted as showing that participants generated their confidence judgments using the metamemory belief that complete recalls are accurate recalls. This metamemory belief was valid for the nondeceptive sentences, which led to the strong positive relationship between confidence and accuracy. The reduced relationship for the deceptive items was interpreted as being due to the participants not being aware of making errors on the deceptive items, and thus using their metamemory belief about completeness to assign inappropriately high confidence to recall errors. Overall the results were taken to support the metamemory theory of confidence.