This study proposes two identification methods for nonlinear rational models. The first is the compound iterative algorithm which is based on the hierarchical technique, which transforms the rational model into two sub-models whose parameters are estimated iteratively using the particle swarm optimization algorithm and intelligent multi-step-length algorithm, respectively. Thus, it has less computational efforts and higher estimation accuracy when compared with the traditional identification methods. Then, a flexible recursive least squares algorithm is proposed which turns the rational model into a time-varying model, thus it does not require the knowledge of the structure of the denominator model, and can update the parameters with new collected data. Simulation results show the effectiveness of the proposed algorithms.