A study was designed to determine whether the presence or absence of a classroom observer and the prior knowledge or lack of knowledge that an observation was to occur would affect the verbal behavior of teachers as measured by the Flanders System of Interaction Analysis. The variables were dichotomized yielding a 2 x 2 x 2 experimental design which allowed observation (through an electronic monitoring system) of two groups of four elementary school teachers (high and low manifest anxiety) under four conditions. The comparisons between teacher's behaviors when an observer was present and their behaviors when no observer was present indicated that teachers become more like their perceived ideal teacher when an observer is present. But it was found that they do not behave more like their perceived ideal teacher when informed of an observation prior to its occurrence than they do when not informed. No support was given to the predicted interaction between a teacher's level of manifest anxiety and "observer present/teacher informed" compared with "observer not present/teacher not informed" conditions. Additional analyses of variance on 41 other interaction variables indicated that the presence or absence of an observer is significantly related to teachers' classroom behavior; changes are in the direction of more indirect behaviors when an observer is present. (Implications of the findings are discussed.) (JS)