Objectives: To compare online cognitive-behavioral therapy (CBT) with and without telephone support respectively to online psychoeducation in a randomized controlled trial (RCT) in caregivers of people with dementia with mild anxiety or depression. Methods: Three-arm parallel-group RCT comparing online CBT with and without telephone support respectively to online psychoeducation. Online study with caregivers of people with dementia. The primary outcome measure was mental health measured by General Health Questionnaire–12 (GHQ-12) at 26 weeks. Secondary outcomes included the Hospital Anxiety and Depression Scale (HADS); the Relative Stress Scale (RSS) and the Short Sense of Competency Questionnaire. The primary analysis focused on people completing GHQ-12 at both baseline and 26 weeks, evaluated using analysis of covariance. Results: 638 people were randomized to the 3 treatment arms, of whom 208 were included in the analysis population. There were significant improvements in GHQ-12 in all treatment arms compared to baseline (P < .001 for all interventions), but neither CBT with nor without telephone support conferred any significant advantage compared to psychoeducation. For the secondary outcomes, there were no significant differences between CBT with telephone support and psychoeducation, but CBT without telephone support was less effective than psychoeducation with respect to HADS depression subscale [mean difference 1.86, 95% confidence interval (CI) 0.61, 3.11; P = .004] and caregiver stress (RSS mean difference 3.11, 95% CI 0.13, 6.09; P = .04). Good safety was achieved in all 3 treatment arms, with no deaths or serious adverse events. Conclusions and Implications: Online CBT with telephone support and psychoeducation both achieved significant benefits over 26 weeks compared with baseline in mental health and mood, but there were no advantages for CBT compared with the psychoeducation intervention. CBT without telephone support was less effective with respect to mood outcomes than psychoeducation and should not be recommended based on current evidence.