The effectiveness of positive and negative evaluative comments for children's learning was assessed in a two-choice discrimination task. Results indicate that negative comments after incorrect responses greatly facilitated learning while positive comments after correct responses had little effect. To explain the findings, a naturalistic analysis of the use of evaluative expressions in the classroom is advanced. It reveals that positive evaluations were used frequently, indiscriminately and often independently of the children's behavior. A distinctive feature of the work is its emphasis upon gains to be made from an integration of traditional experimental methods with ethological analyses of children's behavior. (Author)