Alpha/theta (a/t) neurofeedback training has in the past successfully been used as a complementary therapeutic relaxation technique in the treatment of alcoholism. In spite of positive clinical outcomes, doubts have been cast on the protocol's specificity when compared to alternative relaxation regimes. This study investigated the basic tenet underlying the a/t training rationale, that accurate a/t feedback representation facilitates the generation of these frequency components. Two groups of healthy volunteers were randomly assigned to either (a) real contingent a/t feedback training or (b) a noncontingent mock feedback control condition. The groups were compared on measures of theta/alpha (t/a) ratios within and across training sessions, as well as activational self-report scales after each session. The contingent a/t feedback group displayed significant within-session t/a ratio increments not evident in the mock control group, as well as higher overall t/a ratios in some but not all of the training sessions. No differences were found between the groups in terms of subjective activational phenomenology, in that both groups reported significantly lower levels of activation after training sessions. The data demonstrate that irrespective of considerations of clinical relevance, accurate a/t neurofeedback effectively facilitates production of higher within-session t/a ratios than do noncontingent feedback relaxation.