A study tested two hypotheses: (1) that an inservice training program will alter teacher verbal behavior and (2) that this altered behavior will be associated with increased pupil achievement. Subjects were elementary teachers in two buildings, one the experimental (23) and one the control group (20) and their pupils in grades 1 through 5 (596 experimental and 594 control). Pretests included intelligence tests and Stanford Achievement Tests in arithmetic for students and the Teaching Situations Reaction Test (TSRT) for teachers; teachers were observed (via Flanders' interaction analysis) during six 15-minute sessions of arithmetic instruction. There were no statistically significant differences between the two groups in pupil ability or achievement or teacher behavior. An inservice training program administered to experimental teachers during the semester consisted of 14 hours' instruction in skills to develop a positive social-emotional climate. Included was use of interaction analysis as a feedback technique and emphasis on 14 behavioral variables to be "trained-in" and "trained-out." Ten trained and statistically reliable observers collected interaction analysis intermediate and post data on teacher behavior (186 hours of observation). Data was subjected to t test analysis. Hypothesis 2 was rejected on the basis of achievement posttests. Hypothesis 1 was partially accepted: five of the seven train-in variables accepted, all seven train-out variables rejected. (JS)