The aim of this study was to assess the effectiveness of a cognitive behavioral therapy (CBT) stepped care model (psychoeducation, guided internet treatment, and face-to-face CBT) compared with direct face-to-face (FtF) CBT. Patients with panic disorder or social anxiety disorder were randomized to either stepped care (n = 85) or direct FtF CBT (n = 88). Recovery was defined as meeting two of the following three criteria: loss of diagnosis, below cut-off for self-reported symptoms, and functional improvement. No significant differences in intention-to-treat recovery rates were identified between stepped care (40.0%) and direct FtF CBT (43.2%). The majority of the patients who recovered in the stepped care did so at the less therapist-demanding steps (26/34, 76.5%). Moderate to large within-groups effect sizes were identified at post-treatment and one-year follow-up. The attrition rates were high: 41.2% in the stepped care condition and 27.3% in the direct FtF CBT condition. These findings indicate that the outcome of a stepped care model for anxiety disorders is comparable to that of direct FtF CBT. The rates of improvement at the two less therapist-demanding steps indicate that stepped care models might be useful for increasing patients’ access to evidence-based psychological treatments for anxiety disorders. However, attrition in the stepped care condition was high, and research regarding the factors that can improve adherence should be prioritized.