This　paper　presents　a　musculoskeletal-driven　characterization　of　the　human　arm　endpoint　rotational　stiffness　(HA-ERS)　from　the　perspective　of　biomechanics.　First,　a　3-DOF　wrist　mechanism　is　designed　to　interact　with　subjects,　and　the　perturbation-based　HA-ERS　measurements　are　realized.　Second,　to　eliminate　subjects’　unpredictable　proactive　behavior　in　perturbation　experiments,　the　data　processing　techniques,　including　k-means　method,　probability　distribution,　and　Mahalanobis　distance,　are　adopted　to　cluster,　augment,　and　filter　the　experimental　data,　respectively.　Third,　the　musculoskeletal　behavior　of　the　human　forearm　is　modeled　through　a　Hill-based　musculoskeletal　model　which　incorporates　active　and　passive　muscle　contractions,　muscular　co-activation　effects,　and　muscle-joint　stiffness　relationships.　The　model　is　parameterized　to　account　for　individual　gaps　in　musculoskeletal　dynamics,　including　optimal　muscle　length,　peak　muscle　force,　muscle　activation　levels　(MALs),　etc.　Finally,　the　model＇s　parameters　are　identified　and　the　predictive　ability　is　evaluated.　Experimental　results　show　that　the　proposed　model　can　effectively　estimate　the　HA-ERS,　enabling　a　more　comprehensive　characterization　of　human-compliant　behavior,　and　showcasing　significant　potential　in　human–robot　integration　scenarios.　©　2025　Elsevier　Ltd