Though therapist-guided Internet-based cognitive behavior therapy (ICBT) appears to be efficacious for depression, social anxiety disorder, and panic disorder, relatively little is known about real-world settings and predictors of treatment effects derived from cognitive-behavioral theory. We examined treatment effectiveness and predictors of improvement in a prospective cohort study where patients took part in 10 weeks of ICBT for depression (n = 114), social anxiety disorder (n = 150), or panic disorder (n = 106) at a teaching clinic. Patients self-reported symptoms before, during, and after treatment. Effect sizes were large for improvement in the primary symptom domain of each treatment group: depression (d = 1.48), social anxiety disorder (d = 1.01), and panic disorder (d = 1.15). In ICBT for depression, having no previous experience of psychological treatment (r = 0.21), and more frequent baseline negative automatic thoughts (r = 0.20) predicted larger improvement. In ICBT for panic disorder, having more baseline safety behaviors (r = 0.25) predicted larger improvement. Predictors remained significant when baseline symptoms were included in the statistical models. We conclude that ICBT can be effective in a real-world teaching clinic, and that patients with greater deficits at baseline benefit the most.