Biological removal of nitrogen is a two-step process: aerobic autotrophic microorganisms oxidize ammoniacal nitrogen to nitrate, and the nitrate is further reduced to elementary nitrogen by heterotrophic microorganisms under anoxic condition with concomitant organic carbon removal. Several state variables are involved which render process monitoring a demanding task, as in most biotechnological processes, measurement of primary variables such as microorganism, carbon and nitrogen concentrations is either difficult or expensive. An alternative is to use a process model of reduced order for on-line inference of state variables: based on secondary process measurements, e.g. pH and redox potential. In this work, two modeling approaches were investigated: a generic reduced order model based on the generally accepted IAWQ No. 1 Model [M. Henze, C.P.L., Grady. W., Gujer. G.V.R., Marais. T., Matsuo, Water Res. 21 (5) (1987) 505-515] - generic model(GM), and a reduced order model specially validated with the data acquired from a bench-scale sequential batch reactor (SBR) specific model (SM). Model inaccuracies and measurement errors were compensated for with a Kalman filter structure to develop two state observers: one built with GM, the generic observer (GO), and another based on SM, the specific observer (SO). State variables estimated by GM, SM, GO and SO were compared to experimental data from the SBR unit. GM gave the worst performance while SM predictions presented some model to data mismatch. GO and SO, on the other hand, were both in very good agreement with experimental data showing that filters add robustness against model errors, which reduces the modeling effort while assuring adequate inference of process variables.