The invention discloses a network selection method based on non-cooperative game, so as to adapt to selection of a dynamic network in which network resources and number of accessing users change constantly. The method comprises the following steps of building a non-cooperative game model with maximized system throughput as a target; converting the non-cooperative game model into a common potentialgame model, and defining a potential function in the potential game model as a negative value of weighted interference sum of all users; converting a problem of selecting a network with largest throughput into a problem of selecting a network with minimum weighted interference sum; and resolving Nash equilibrium of the potential game model by using a distributed learning algorithm, wherein the acquired solution is an optimal access network selection solution. Through adoption of the method, the system throughput is maximized, the calculation complexity is lowered effectively through the learning algorithm, and the system throughput of the convergent solution is approximate to the optimal system throughput.