In this paper, we address the problem of structure identification of Volterra models. It consists in estimating the model order and the memory lcngth of each kernel. Two methods based on input-output crosscumulants arc developed. The first one uses zero mean independent and identically distributed Ganssian input, and the second one concerns a symmetric input sequence. Simulations are performed on six models having different orders and kernel memory lengths to demonstrate the advantages of the proposed methods.