In this study, we aimed to assess the factors that predict a dysfunctional conditioned pain modulation (CPM) in chronic knee OA. This is a cross-sectional analysis of patients with chronic knee OA from a prospective cohort study in Brazil (n = 85). We performed linear and logistic multivariate regression models using the purposeful selection approach to test the relationship between the CPM in both knees (average) as a dependent variable and demographics, clinical, and neurophysiological as independent variables. A significant negative association between WOMAC pain scores and CPM (β: -0.13) was found. This association was modified by the subjects’ race, being stronger in the non-white subjects. In our logistic regression models, pain intensity indexed with the WOMAC pain scale remained a significant association with dichotomized CPM. Furthermore, a significant CPM association with balance, indexed with the Berg Balance score, was evidenced (β: 0.04). Neurophysiological variables showed a significant negative relationship with CPM, such as the relative power of delta oscillations in the frontal area (β: -3.11) and central area (β: -3.23). There was no significant relationship between CPM and the following domains: cognitive, emotion, sleep, opioid receptor polymorphisms, and intrinsic variables of OA disease. There was no association of CPM with TMS-indexed inhibitory markers. These results may indicate that less function of the pain descending inhibitory system in patients with OA is correlated with higher activity-related pain (WOMAC), less balance, and cortical plasticity especially with increased low-frequency (delta) brain oscillations. These associations seem modified by race.