One of the most appealing features of robot-assisted-therapy (RMT) consists in the possibility to quantify the performances of the rehabilitation task proposed to the patient. The analysis of kinematics indexes, providing quantitative information about movement patterns that is not captured using clinical measures, could be used as reliable outcome measures in clinical upper limb rehabilitative settings. Few papers have actually addressed this issue and aim of the study is the assessment of the reproducibility of kinematics indexes of upper arm horizontal reaching movements in normal subjects during RMT. We studied two repeated sets of 60 reaching movements of 25° amplitude at 20-30 and 40°/s target velocity in 5 healthy subjects. Reaching movements were automatically detected and quantified by symmetry, skewness, kurtosis and smoothness indexes. Results showed agreements' limits of kurtosis and smoothness narrower than other kinematics indexes, with Bland-Altman plots rapidly increasing their scatter with higher values. This suggesting to limit their evaluation only to the average of a sequence of medium velocity reaching movements of about 30°/s.