The present study examined the depressive realism hypothesis, which posits that depressed people are often more accurate in their perceptions and judgments than non-depressed people are. Each subject initially generated descriptions of others important to them, both positive and negative important others, in the pretest section, and 3 or 4 weeks later, some subjects were invited into the formal experiment to measure the accuracy of their perceptions in a pseudo-social interaction situation. A total of 52 patients diagnosed with clinical depression and 62 normal matched subjects participated in the experimental procedure. The results indicate that clinically depressed patients provided more accurate, less distorted descriptions of their positive important others than did those in the normal group. However, when information involved the negative important others, the results exhibited a trend, but these results did not provide significant support for the depressive realism hypothesis. The results support the depressive realism hypothesis when tasks involve subjects' own positive important others.