The present study examined the effects of writing a composition on a haiku upon comprehension of the haiku. Subjects were devided into two groups (Writing-Before group and Writing-After group). The task sequence of the Writing-Before group was as follows: (1) writing a composition on a haiku, (2) the first impression rating of the haiku (semantic differential technique), (3) evaluation of a composition written by a stimulus person, (4) inferring the impression rating of the stimulus person on the basis of his composition on the haiku, and (5) the second impression rating of the haiku. The task sequence of the Writing-After group was the same as that of Writing-Before group except that the subjects wrote a composition at the end of the sequence. The results indicated that the Writing-Before group inferred the impression ratings of the stimulus person more accurately than the Writing-After group and the discrepancy of impression ratings between Writing-Before group and stimulus person decreased in the second impression rating. Theoretical implications of these findings were discussed.