ABSTRACT Background Sarcopenia is a devastating disease for older adults, but it lacks accessible and reliable tools for measuring total appendicular skeletal muscle mass (ASMM). Two‐dimensional muscle ultrasound (US) has been developed for its bedside clinical advantages and feasibility but lacks standardization and prediction performance. We previously validated a new 3D‐US technique to measure muscle volume (MV) at bedside and applied it in a geriatric rehabilitation setting. Objectives were to analyse the concordance between 3D‐US MV and ASMM and compare concordance between 3D‐US MV and 2D‐US parameters with ASMM. Methods Participants were recruited in a Geriatric rehabilitation ward in Nantes, France, from May to October 2022. Exclusion criteria were as follows: oedema in the lower limbs or recent history of unilateral lower limb damage or stroke. ASMM was measured with bioelectrical impedance analysis; 3D‐US and 2D‐US acquisitions were performed on three muscles of the right lower limb. Measures of strength (hand grip, knee extension and ankle dorsiflexion) were also recorded. Reliability of 3D‐US MV measurements on 10 participants was high (ICC = 0.99). We used Lin's concordance correlation coefficients (CCC) and bias correction factor for agreement between variables and linear regression models for prediction equations. Results Fifty‐eight participants had an interpretable ASMM of whom 17 (29%) had a diagnosis of sarcopenia. Volumes of TA, RF and VL were all significantly concordant with ASMM measured by BIA (all p values < 0.001), with CCCs respectively of 0.72, 0.61 and 0.60. MV were all significantly concordant with isometric strength ( p values < 0.001). Concordance and correlation with ASMM were higher with 3D‐US than 2D‐US measurements regardless of the muscle. Prediction of ASMM reached an adjusted R 2 of 0.8 with tibialis anterior volume, biometrics and 2D measurements. Conclusions This study was the first to use 3D‐US in a geriatric setting and develop a model to predict ASMM in very old hospitalized patients. MV measurements with 3D‐US proved to be reliable and more concordant with appendicular muscle mass and strength than 2D parameters.