This paper considers the parameter estimation problem of a water level system described by a pseudo-linear autoregressive moving average(ARMA)model.We derive the state space model of a three-tank water level system by using the fluid dynamics,and obtain the discrete-time model and its input-output representation by using the linearization method and the Euler method.By using the filtering technique,a filtering based gradient iterative(F-GI)algorithm is given and a filtering based least squares iterative(LSI)algorithm is presented for improving the convergence rates and furthermore,a filtering based decomposition LSI algorithm is derived for improving the computational efficiency.The key is to use the hierarchical identification principle,to apply the data filtering technique and the decomposition technique to identification,and to replace the unknown terms in the information vectors with their estimates.Compared with the F-GI algorithm,the filtering based LSI algorithm and the filtering based decomposition LSI algorithm have faster convergence rates.The simulation results indicate that the proposed algorithms are effective.