Responsiveness of clinical assessments is an important element in the report of clinical effectiveness after rehabilitation. The correlation could reflect the validity of assessments as an indication of clinical performance before and after interventions. This study investigated the correlation and responsiveness of Fugl-Meyer Assessment (FMA), Motor Status Scale (MSS), Action Research Arm Test (ARAT) and the Modified Ashworth Scale (MAS), which are used frequently in effectiveness studies of robotic upper-extremity training in stroke rehabilitation. Twenty-seven chronic stroke patients were recruited for a 20-session upper-extremity rehabilitation robotic training program. This was a rater-blinded randomized controlled trial. All participants were evaluated with FMA, MSS, ARAT, MAS, and Functional Independent Measure before and after robotic training. Spearman's rank correlation coefficient was applied for the analysis of correlation. The standardized response mean (SRM) and Guyatt's responsiveness index (GRI) were used to analyze responsiveness. Spearman's correlation coefficient showed a significantly high correlation (rho = 0.91-0.96) among FMA, MSS, and ARAT and a fair-to-moderate correlation (rho = 0.40-0.62) between MAS and the other assessments. FMA, MSS, and MAS on the wrist showed higher responsiveness (SRM = 0.85-0.98, GRI = 1.59-3.62), whereas ARAT showed relatively less responsiveness (SRM = 0.22, GRI=0.81). The results showed that FMA or MSS would be the best choice for evaluating the functional improvement in stroke studies on robotic upper-extremity training with high responsiveness and good correlation with ARAT. MAS could be used separately to evaluate the spasticity changes after intervention in terms of high responsiveness.