Data-driven discovery of governing equations is revolutionizing research fields where scientific data are abundant but well-characterized quantitative descriptions remain scarce. In this work, we combine molecular simulations with symbolic regression to discover governing equations for fluid dynamics. Molecular simulations serve as the data source, offering the key advantage of relying solely on microscopic models of molecular interactions, without assuming macroscopic governing equations a priori. To extract governing laws from such data, we propose a general Symbolic Identification of Tensor Equations (SITE) framework. SITE is capable of discovering both scalar and tensor forms of governing equations, while ensuring strong physical interpretability through dimensional homogeneity check, and achieving efficient coefficient optimization via tensor linear regression. The framework has been successfully applied to identify constitutive relations and governing equations across diverse scenarios, including incompressible and compressible flows, and even non-equilibrium states.