Test anxiety is widespread and associated with poor performance in academic examinations. The Internet, not well-proven for the treatment of anxiety, should be able to deliver highly accessible Cognitive Behavior Therapy (CBT). This study sought to test the hypothesis that CBT, available on the Internet, could reduce test anxiety. Ninety university students were randomly allocated to CBT or a control program, both on the Internet. Before and after treatment, the participants completed the Test Anxiety Inventory (TAI), an Anxiety Hierarchy Questionnaire (AHQ), the Exam Problem-Solving Inventory (EPSI), the General Self-Efficacy Scale (GSES) and the Heim reasoning tests (AH) as a measure of test performance. Of the CBT and control groups 28% and 35%, respectively, withdrew. According to the TAI, 53% of the CBT group showed a reliable and clinically significant improvement with treatment but only 29% of the control group exhibited such a change. On the AHQ, 67% of the CBT group and 36% of the control group showed a clinically significant improvement, more than two standard deviations above the mean of the baseline, a change in favour of CBT. Both groups improved on the GSES, in state anxiety during exams retrospectively assessed, and on the AH tests. The improvement on the AH tests was probably a practice effect and not a reflection of a change in capacity for academic testing. This study thus supports use of CBT on the Internet for the treatment of test anxiety.