The fundamental way out for agricultural development is to improve total factor productivity and to expand its contribution to agricultural growth. The Malquist index method was used to calculate the agricultural total factor productivity of each province from 1998 to 2015. The multilevel linear model was used to empirically analyze its influencing factors from the perspective of social embeddedness. The results show that among the overall differences in agricultural total factor productivity of each province,36.46% is caused by the difference in its social embeddedness; agricultural structural coefficient,urbanization process,disaster rate,rural human capital,and logistics relationship embedding have direct impacts on agricultural total factor productivity,yet cognitive relationship embedding,capital quantity embedding,structural relationship embedding,cultural relationship embedding,and political relationship embedding have indirect effects,which can be achieved by adjusting the relationship between agricultural structural coefficients,urbanization processes,and rural human capital and agricultural total factor productivity. Therefore,to improve the total factor productivity of agriculture,it is necessary to consider both the basic influencing factors and social embedded factors.