Objective An evaluation system of upper limb movement function based on microelectro mechical system (MEMS) accelerometer sensor is designed to accurately and objectively evaluate the dysfunction of upper limb movement in stroke patients through the index of smoothness and root mean square of acceleration. Methods The system mainly included ATmega128 single-chip microcomputer communication module, this module could transmit data between the upper and lower computer and save the data. The system also included data acquisition module, this module could collect the patient's upper limb motion acceleration in real time by using ADXL345 acceleration sensor. The system was low cost, easy to carry and efficient in tests.Finally, six stroke patients with upper limb hemiplegia rehabilitation (patient group) and four normal healthy adults (healthy group) were recruited. Healthy group performed directional motion with zero load and reduplicate tests of directional motion with gradient load.The data was contrasted between this two groups when patient group and the healthy group were tested of directional motion with zero load. Results Repeat tests of oriented motion with zero load: there were no significant differences among the smoothness (P> 0. 05) and the mean square root of acceleration (P> 0. 05) between numbers in the healthy group. Contrast test of oriented motion with zero load: there were significant differences among the smoothness (P < 0. 05) and no significant differences among the mean square root of acceleration (P> 0. 05) between two groups. Repeat tests of directional motion with gradient load: smoothness increased as the load increased. And the smoothness (P< 0. 001) was significantly linear related to the load by the unitary linear regression analysis. Conclusions The evaluation system of upper limb movement function based on MEMS acceleration sensor can quantitatively evaluate the movement capacity of upper limb in stroke patients through the index of smoothness. Furthermore, the linear relationship between smoothness and load is obtained and provided a new way to evaluate the recovery of stroke patients with upper limb hemiplegia.