BACKGROUND: Generalized anxiety disorder (GAD) is associated with the lowest treatment response rate among all anxiety disorders. Understanding mechanisms of improvement may help to develop more effective and personalized treatments. AIM: The objective of the study was to investigate different improvement mechanisms in the treatment of individuals diagnosed with GAD. DESIGN: We reported data from a randomized controlled trial that evaluated three different GAD treatments (mindfulness-based intervention, BMT; fluoxetine, FLX; and an active comparison group, QoL) for 8 weeks. METHOD: Mediation analyses were performed evaluating the association between worry symptoms at baseline and anxiety scoring at the endpoint, considering self-compassion or mindfulness or its dimensions at mid-treatment as mediators for the whole sample (assessing GAD improvement mechanism) and the different interventions as moderators. RESULTS: Contrary to mindfulness state scoring (C = .06; 95% CI = -.05 to .20), self-compassion (C = .11; 95% CI = .01 to .28) and non-judgement of inner experience (C = .10; 95% CI = .004 to .21) mediated the association between worry symptoms at baseline and anxiety at the endpoint. When comparing BMT to FLX, the intervention modality did not moderate these associations. CONCLUSION: Self-compassion and non-judgement of inner experience seem to be essential targets in GAD treatment, contrary to the mindfulness state itself. Although no difference was found considering the intervention modality, future research may assess how to boost these dimensions in specific treatments for GAD.