This paper reports the first nonlinear dynamical model of human hippocampus for building memory prostheses. In this study, spike trains are recorded from the hippocampal CA3 and CA1 regions in epileptic patients performing two memory-dependent behavioral tasks. Using CA3 and CA1 spike trains as inputs and outputs respectively, second-order generalized Laguerre-Volterra models are built to capture the nonlinear dynamics underlying the spike train transformations. These models can predict the CA1 spike trains based on the CA3 spike train and thus be used as the computational basis of the hippocampal memory prosthesis.