Objective The objective of this study was the measurement of the test-retest reliability of n-back in Chinese stroke patients. Methods Seventy-five sub-acute stroke patients performed n-back twice in three days. The test-retest reliability of n-back was analyzed by correlation coefficient. Results The n-back had excellent test-retest reliability in stroke patients. Pearson or Spearman coefficients ranged from 0.81 to 0.88. The intra-class correlation coefficients ranged from 0.72 to 0.87. The Chinese version of Montreal Cognitive Assessment-Basic (MoCA-BC) score was significantly correlated with the performance of n-back. MoCA-BC and n-back accuracy were significantly related in the Mild Cognitive Impairment (MCI) group (r = 0.60 in 1-back, p = .002; r = 0.43 in 2-back, p = .040). However, MoCA-BC was correlated with reaction time (RT) in the Cognitively Normal (CN) group (r = -0.44 in 1-back, p = .003; r = -0.36 in 2-back, p = .018). The test-retest reliability of CN group was mostly higher than that of MCI group RT: 0.71-0.76 in MCI, 0.80-0.88 in CN; accuracy: 0.80-0.85 in MCI, 0.75-0.86 in CN). The practice effect was observed in the CN group instead of the MCI group. Conclusions This study indicated that the test-retest reliability of n-back was high in stroke patients. N-back was correlated with cognition. It was preferable to conduct subgroup analyses according to the level of cognitive assessment of patients with stroke.