Knee osteoarthritis (OA) is the most common musculoskeletal disorder affecting all populations. One common knee OA symptom is instability; thus its assessment could allow diagnosing and following-up of the disease without using conventional imaging techniques, such as plain radiography or magnetic resonance imaging (MRI). Knee kinematic measurements using accelerometers could provide a low-cost and non-invasive option to quantify knee instability. The aim of this study was to assess the relationships between kinematic data, instability parameters derived from the imaging techniques, goniometer-based measurements, and radiological OA stage. The right knees of 66 females (44-67 years) were examined using MRI, plain radiography, and goniometer-based angle measurement. Kellgren-Lawrence (KL) grade and the joint line convergence angle (JLCA) were determined from the radiographs. Cartilage thickness and OA score (MOAKS) were derived from the MRI. A ratio between lateral and medial cartilage thicknesses was calculated from the average thickness of segmented cartilage over the weight bearing area (MRIratio). Accelerometers attached to thigh and shank were used to record kinematic signals during a one-leg-stand test. Power of the accelerometer signals along the anatomical longitudinal axis (Pacc) was used as a measure of knee instability. Finally, Spearman's correlations between the acquired parameters and KL grade / MOAKS scores were calculated. Leave-one-out cross-validation and logistic regression were used to discriminate OA subjects (KL ≥ 2). All the instability parameters (Pacc, JLCA and MRIratio), except the goniometer angle, showed significant correlations with KL grading (rho=0.32-0.64, p<0.01) and MOAKS composite score (rho=0.35-0.56, p<0.01). Both Pacc and JLCA showed higher areas under the ROC curve to discriminate OA (AUC=0.76 and AUC=0.78) than MRIratio and goniometer angle (AUC=0.55 and AUC=0.56). Our results demonstrate the clinical potential of kinematic knee instability measurements using low-cost accelerometers. Such approach could become a potential new tool in OA diagnostics.