Heart rate variability (HRV) can be conceptualized as a marker of an individual's capability to adaptively respond to its environment and has been linked with mental health. Although conceptually and empirically linked to social behavior and thus relevant in the therapeutic setting, HRV is seldom investigated directly within therapy sessions. In the present examination, we aimed at addressing this research gap by assessing patients' and therapists' HRVs both ambulatory within therapy sessions and under resting conditions. Drawing on polyvagal theory, we hypothesized that higher in-session HRV is accompanied with higher therapeutic alliance ratings. Further, we expected baseline HRV to predict symptomatic outcome and to increase over the course of therapy. In a sample of 53 outpatients receiving 25 sessions of cognitive behavioral therapy, we measured HRV, therapeutic alliance and depressive symptoms on four occasions. Multilevel modeling analyses demonstrated that patients with higher in-session high-frequency HRV rated the therapeutic alliance higher. Baseline HRV predicted symptomatic outcome and increased over the course of therapy. Possible explanations involve a link between in-session HRV and in-session behavior and should be investigated in future studies. The results highlight the usefulness of in-session HRV as a promising process variable in psychotherapy research.