Upper-limb functional assessment is important for stroke treatment. The identification of sensitive kinematic metrics that best differentiate the impairment level of upper-limb motor function can enhance this assessment. Therefore, this research proposed a method to select sensitive kinematic metrics which can discriminate between stroke patients and healthy subjects. A total of 26 participants (10 healthy subjects and 16 stroke patients) were recruited to perform upper-limb reaching movements. The movement data was measured using Kinect v2. Thirty-two metrics were then extracted. Independent samples T-test, Mann-Whitney U-test and principal component analysis were performed to select sensitive metrics. Experimental results show that the first principal component explained 54.67% of the total variance, and it can distinguish stroke patients from healthy subjects. Meanwhile, loading values of index of curvature and spectral arc-length were 0.895 and 0.831 respectively, which contributed most for the first principal component. Therefore, we concluded that the sensitive metrics were index of curvature and spectral arc-length, which had significant importance to differentiate stroke patients from healthy subjects.