A pre/post pilot study was designed to investigate neurobiological mechanisms and plasma metabolites in an 8-week Tai-Chi (TC) group intervention in subjects with knee osteoarthritis. Twelve postmenopausal women underwent Tai-Chi group exercise for 8 weeks (60 min/session, three times/week). Outcomes were measured before and after Tai Chi intervention including pain intensity (VAS), Brief Pain Inventory (BPI), Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), plasma metabolites (amino acids and lipids), as well as resting-state functional magnetic resonance imaging (rs-fMRI, 10 min, eyes open), diffusion tensor imaging (DTI, 12 min), and structural MRI (4.5 min) in a subgroup. Clinical data was analyzed using paired -tests; plasma metabolites were analyzed using Wilcoxon signed-rank tests; and rs-fMRI data were analyzed using seed-based correlations of the left and right amygdala in a two-level mixed-effects model (FSL software). Correlations between amygdala-medial prefrontal cortex (mPFC) connectivity and corresponding changes in clinical outcomes were examined. DTI connectivity of each amygdala was modeled using a Bayesian approach and probabilistic tractography. The associations between neurobiological effects and pain/physical function were examined. Significant pre/post changes were observed with reduced knee pain (VAS with most pain: = 0.018; WOMAC-pain: = 0.021; BPI with worst level: = 0.018) and stiffness (WOMAC-stiffness, = 0.020), that likely contributed to improved physical function (WOMAC-physical function: = 0.018) with TC. Moderate to large effect sizes pre/post increase in rs-fMRI connectivity were observed between bilateral mPFC and the amygdala seed regions (i.e., left: = 0.988, = 0.355; right: = 0.600, = 0.282). Increased DTI connectivity was observed between bilateral mPFC and left amygdala ( = 0.720, = 0.156). There were moderate-high correlations ( = 0.28-0.60) between TC-associated pre-post changes in amygdala-mPFC functional connectivity and pain/physical function improvement. Significantly higher levels of lysophosphatidylcholines were observed after TC but lower levels of some essential amino acids. Amino acid levels (alanine, lysine, and methionine) were lower after 8 weeks of TC and many of the lipid metabolites were higher after TC. Further, plasma non-HDL cholesterol levels were lower after TC. This pilot study showed moderate to large effect sizes, suggesting an important role that cortico-amygdala interactions related to TC have on pain and physical function in subjects with knee osteoarthritis pain. Metabolite analyses revealed a metabolic shift of higher lyso-lipids and lower amino acids that might suggest greater fatty acid catabolism, protein turnover and changes in lipid redistribution in response to TC exercise. The results also support therapeutic strategies aimed at strengthening functional and structural connectivity between the mPFC and the amygdala. Controlled clinical trials are warranted to confirm these observed preliminary effects.