[Purpose] This pilot study aimed to investigate the relative and absolute reliability of variables obtained from an acceleration-based gait analysis conducted at comfortable and maximal gait speeds in individuals with chronic stroke. [Participants and Methods] This study included 25 community-dwelling individuals with chronic stroke. The participants wore triaxial accelerometers, while an observed walking trial was performed at comfortable and maximal speeds on two separate days 1 week apart. Relative reliability was evaluated using the intraclass correlation coefficient, and absolute reliability was evaluated using the Bland–Altman analysis, standard error of measurement, and minimal detectable change. [Results] The intraclass correlation coefficient of gait varied according to the acceleration-based gait analysis, ranging from 0.70 to 0.99. The Bland–Altman analysis revealed no systematic bias in both comfortable and maximal gait speed conditions. Most of the minimal detectable changes were smaller at maximal gait speed than at comfortable gait speed. [Conclusion] Acceleration-based gait analysis is a reliable method, particularly in maximal gait speed conditions. It may be used to assess the effect of rehabilitation interventions in individuals with chronic stroke.