Marrow extraction has been an important component of butchery practice in Paleolithic China. The diagnostic of this kind of human behavior could not be based on isolated evidence,but on multiple clues,such as the outline,texture and angle of long bone fracture,the circumference of mid-shaft fragment as well as the morphology,frequency and placement of percussive damages(percussion marks and notches). In this paper,compared to the actualistic(experimental and ethnoarchaeological)studies on marrow harvesting,the Ma'anshan long bone assemblage is analyzed to demonstrate the characteristics of the bone fragments derived from marrow extraction by prehistoric humans. Ma'anshan site (28°07'18''N, 106°49'37''E) is located 2km southeast of Tongzi County,northwest Guizhou Province. It is a cave site that lies at an altitude of 960m above sea level and 40m higher above the nearby Tianmen River. This cave was systematically excavated in 1986 and 1990 and the cultural deposit is divided into two layers. The upper layer dates between 15kaB.P.and 31kaB.P.and the lower layer is around 53kaB.P. Along with some stone and bone artifacts,more than ten thousand of bone fragments from fifteen genera of large mammals were recovered from the cave deposits. The outline,texture and angle of the fracture plane of the bone fragments from the cave is generally similar to those of bones experimentally fragmented when they are fresh or green. However,since some other taphonomic agents,for example carnivores,could also break bones in this manner,it's still inconclusive to suggest a marrow-extracting behavior for humans in this cave based solely on this thread of evidence. A predominant percentage(85%)of the long bones from this faunal assemblage with circumference less than or equal to one fourth of their original states,which is in good accordance with that of the animal bones fragmented by experimenters in actualistic studies. Percussion marks and notches are newly proposed for the identification of human marrow-extraction behaviors from archaeological records. No percussion marks were observed on bone surfaces from the Ma'anshan site,probably due to the somewhat large thickness of the periosteum or the smoothness of the stone tools employed by humans to break bones,the moderately damaged surfaces from root-etching may also contribute to this fact. Percussion notches,on the contrary,are well recorded from this faunal assemblage. In contrast to the much higher percentages of near-epiphyses notched by Paleolithic humans at this site,only 8.33%~16.67% long bone shafts of large-sized animals bear this category of modifications. This may therefore indicate that humans at the Ma'anshan site,behaved similar to modern hunter-gatherers in Africa in this regard,preferentially impacted at the epiphyseal ends of the long bone to exploit the nutrition-rich marrow within the cavities.