F rom the perspective of achieving an enhanced and superior closed-loop performance using a model based control scheme, accurate estimation of a plant model is the most important step. Towards this end, empirical model identifi cation methods, that use plant data, obtained after appropriate excitation of the plant, have been recommended. Typically, the excitation and model building has been proposed in open loop schemes. However, for open loop unstable plants as well as in situations involving time varying plant characteristics, excitation and model building are also proposed in closed-loop schemes. The added advantage of using closed-loop data for identifi cation is that the data in general, implicitly refl ects the intended operating condition of the plant, and may thus yield a better plant model than that obtained from open loop identifi cation. Irrespective of whether identifi cation is carried out in open loop or closed loop schemes, the infl uence of the model order on the model accuracy in terms of the bias and variance errors is an important consideration. This is particularly relevant when a parsimonious model and ease of controller structure and design is sought, subsequent to the identifi cation process. The task of obtaining a parsimonious yet accurate model requires a compromise between achieving a good match at certain control-relevant frequencies versus large mismatch at other frequencies and has generated considerable interest under the area of joint identifi cation and control or controlrelevant identifi cation.Often times, operating plants cannot afford the expensive task of open loop identifi cation as this violates the primary operating objective. Furthermore, in the presence of nonlinearities, time varying behaviour or changes in the operating regions, the model needs to be updated or adapted under closed-loop conditions. have proposed various approaches towards model identifi cation under closed-loop conditions. Vishwanathan and Rangaiah (2000) have proposed an optimizationbased approach to estimate a second order plus dead time model of the process using closed-loop response. Zang et al. (1992) have proposed an online refi nement methodology for controllers, for disturbance rejection using closed-loop modeling. Schrama and co-workers (1993,1992a, 1992b) have proposed an iterative identifi cation and control design scheme for adaptive performance enhancement using closed-loop identifi cation in the frequency domain with co-prime factor perturbations. Various other techniques have been proposed for closed-loop identifi cation from the point of view of bias and variance error minimization (Forssell and Ljung, 1999;Gevers et al., 2001;Gilson and Van den Hof, 2001). Another area of interest wherein closed-loop identifi cation is useful, is direct controller order reduction proposed by Landau et al. (2001) and Rivera and Morari (1992a). Wang et al. (2001) have proposed a robust closedloop identifi cation scheme with application to auto tuning. Lakshminarayanan et al. (2001a) have used the canoni...