The triple networks, namely the default-mode network (DMN), the central executive network (CEN), and the salience network (SN), play crucial roles in disorders of the brain, as well as in basic neuroscientific processes such as mindfulness. However, currently, there is no consensus on the underlying functional features of the triple networks associated with mindfulness. In this study, we tested the hypothesis that (a) the partial regression coefficient (i.e., slope): from the SN to the DMN, mediated by the CEN, would be one of the potential mindfulness features in the real-time functional magnetic resonance imaging (rtfMRI) neurofeedback (NF) setting, and (b) this slope level may be enhanced by rtfMRI-NF training. Sixty healthy mindfulness-naïve males participated in an MRI session consisting of two non-rtfMRI-runs, followed by two rtfMRI-NF runs and one transfer run. Once the regions-of-interest of each of the triple networks were defined using the non-rtfMRI-runs, the slope level was calculated by mediation analysis and used as neurofeedback information, in the form of a thermometer bar, to assist with participant mindfulness during the rtfMRI-NF runs. The participants were asked to increase the level of the thermometer bar while deploying a mindfulness strategy, which consisted of focusing attention on the physical sensations of breathing. rtfMRI-NF training was conducted as part of a randomized controlled trial design, in which participants were randomly assigned to either an experimental group or a control group. The participants in the experimental group received contingent neurofeedback information, which was obtained from their own brain signals, whereas the participants in the control group received non-contingent neurofeedback information that originated from matched participants in the experimental group. Our results indicated that the slope level from the SN to the DMN, mediated by the CEN, was associated with mindfulness score (rtfMRI-NF runs: r = 0.53, p = 0.007; p-value was corrected from 10,000 random permutations) and with task-performance feedback score (rtfMRI-NF run: r = 0.61, p = 0.001) in the experimental group only. In addition, during the rtfMRI-NF runs the level of the partial regression coefficient feature was substantially increased in the experimental group compared to the control group (p < 0.05 from the paired t-test; the p-value was corrected from 10,000 random permutations). To the best of our knowledge, this is the first study to demonstrate a partial regression coefficient feature of mindfulness in the rtfMRI-NF setting obtained by triple network mediation analysis, as well as the possibility of enhancement of the partial regression coefficient feature by rtfMRI-NF training.