This study was designed to evaluate the effect of semantic priming with a lexical decision task in 22 depressed patients (DSM-III-R, 1987) and 30 control subjects. These patients were evaluated twice: first when they arrived at the hospital, and secondly, after clinical improvement. Clinical improvement was evaluated using standard depression rating scales. A lexical decision task involving semantic relations (related vs. unrelated, e.g., apple-pear) was used to evaluate the processing of semantic information. The results showed that, for the first evaluation, the depressives presented similar semantic priming to control subjects. When we compared semantic priming in the first and the second passes, we observed that its amplitude was identical. The sole difference between the two passes concerns the global reaction time in the depressive group. This last result suggested that, with clinical improvement, the characteristic psychomotor retardation declines. One of the major results concerns the fact that severe depressive patients (first pass) exhibit normal semantic priming in a lexical decision task. These results indicate, in this clinical population, the preservation of controlled processes implicated in this lexical decision task.