[Purpose] This study aimed to verify the appropriate number of measurements and the intrarater reliabilities of shoulder joint horizontal adductor muscle strength measurements using a handheld dynamometer (HHD) for geriatric and stroke patients. [Subjects and Methods] The subjects were 40 inpatients, who were divided into two groups: 20 stroke patients in the stroke group (SG), and 20 geriatric patients in the no-stroke group (N-SG). Measurements were performed three times using an HHD with a belt. The reliability was verified using Bland-Altman analysis and the intraclass correlation coefficient (ICC). [Results] ICC (1, 1) was >0.9. A systematic bias was not observed between the first and second measurement values except for the right side in N-SG. A systematic bias between the maximum value obtained during the first and second measurements and third measurement value was observed on the left side in N-SG, and on the non-paralyzed side in SG: the third measurement values were small in both cases. [Conclusion] Intrarater reliabilities were high for shoulder horizontal adductor strength measurements using an HHD with a belt for geriatric and stroke patients. Taking the systematic bias into consideration, these findings suggest that the required number of measurements is two.