Summary: In this paper, based on artificial neural networks, the problem of function approximation is studied. Further, the convergence for function approximation is analyzed, which is affected by the structural parameters of neural networks. The results indicate that function approximation can be worked out correctly by virtue of artificial neural networks.