Upper limb deficits are common sequelae after a stroke and negatively affect daily living and quality of life. The use of outcome measures to evaluate upper limb function is essential to assess sensorimotor recovery and to determine the effectiveness of rehabilitation. The aim of this study was to estimate the construct validity and inter-rater reliability of three shortened versions of the Singapore version of the Chedoke Arm and Hand Activity Inventory (CAHAI-SG) comprising seven, eight, and nine test items. The sample consisted of 55 inpatients with acute/subacute stroke to whom the CAHAI-SG, Fugl-Meyer Assessment of Upper Extremity (FMA-UE) and the Action Research Arm Test (ARAT) were administered. To estimate convergent and discriminative construct validity, Spearman’s rank correlation coefficient and 95% confidence intervals were computed for CAHAI-SG scores with FMA-UE and ARAT scores. Reliability was estimated using intraclass correlation coefficient (relative reliability) and the standard error of measurement (absolute reliability). Convergent validity with the FMA-UE was 0.79, 0.80, and 0.81 for seven-item, eight-item, and nine-item versions of the CAHAI-SG, respectively, and 0.81 with the ARAT for all shortened versions. Discriminative validity with the FMA-UE pain subscale was between 0.37 and 0.38. The absolute reliability was 3.09, 3.65, and 3.98, and relative reliability was 0.96, 0.95, and 0.96 for the seven-item, eight-item, and nine-item versions, respectively. All shortened versions of the CAHAI-SG demonstrated similar psychometric properties to the full (13 item) version, meaning clinicians may use these shorter versions that require less time to administer and score.