Elman networks' dynamical modeling capability is discussed in this paper firstly. According to Elman networks' unique structure ,a weight training algorithm is designed and a nonlinear adaptive controller is constructed. Without the PE presumption, neural networks controller's closed-loop properties are studied and the whole Elman networks' passivity is demonstrated. § 1 IntroductionAfter nearly twenty yearys' evolution,rich achievements on neural networks control (NNC) have been gained in nonlinear control fields [1]. Based on the multi-layered feed-forward neural networks (MFNN) and the full-connected recurrent neural networks (FRNN),researchers have developed many kinds of new algorithms and viewpoints as well. Because MFNN is essentially a kind of static mapping neural networks ,it can't realize a perfect description to a nonlinear system's dynamics whereas FRNN can. So in NNC, FRNN has its unique advantage over MFNN. However,the training of FRNN weight values is much more complicated[2] than that of MFNN,therefore this restriction can't bring FRNN's whole advantage into play in modeling system dynamics and real-time control. In recent years,finding a kind of simple recurrent networks model to fit the nonlinear control has become a new research hotspot[3'4].Elman networks[5] unique charmhas attacted lots of researchers' attentions, and wide slope researches have been developed in aspects such as nonlinear modeling,temporal signal process and automata theory [6][7][8]. In [6] the dynamic backpropagation algorithm (DBP) was applied to trainElman networks' weight values for modeling nonlinear dynamic systems. In [7] it was shown that Elman networks can simulate any frontier-to-root