This study investigated the cognitions of low, moderate, and high test-anxious children under naturalistic test-taking conditions. As predicted, high test-anxious children showed more task-debilitating cognitions during testing, including more negative self-evaluations and off-task thoughts and fewer positive self-evaluations. High test-anxious children also showed relatively high frequencies of on-task thoughts and coping self-statements. The study also examined the role of classroom environment in the test anxiety-performance relation. Relative to their peers, the performance of high test-anxious children was debilitated only in classrooms that were high in perceived evaluative threat. Moreover, significant relations between cognitions and math performance were obtained only in highthreat classrooms, and these relations were maintained when the effects of math ability were statistically controlled. Future test-anxiety research and clinical intervention might view test anxiety within a broader theoretical context of person-environment fit.The debilitating effects of test anxiety on task performance have been well documented (see Sarason, 1980) and have made test anxiety a topic of substantial clinical interest. As a group, high test-anxious (HA) individuals perform more poorly on laboratory tasks, intelligence tests, achievement tests, and classroom examinations (e.g., Deffenbacher, 1980;Dusek, 1980;Wine, 1980). This study examined the relation of test anxiety and cognitions during testing to children's actual test performance within different classroom environments.Previous research has suggested that cognitions during testing might affect test performance. Wine (1971) proposed that HA individuals divide their attention betweenThe present study is based on a doctoral dissertation submitted by the first author under the supervision of the second author. Appreciation is expressed to the Tempe, Arizona, school district for participation in the study and to Kristin Ihhg for assistance in data collection. Special thanks go to Irwin Sandier, Sharlene Wolchik, George Chartier, and Susan Somerville for their guidance; and to Clark C. Presson for his helpful comments. Appreciation is also expressed to Marsha Stephens and the Division for Disorder of Development and Learning, University of North Carolina-Chapel Hill, for use of their computer facilities.