Estimation of biomass, substrate, and specific growth rate (μ) by two nonlinear observers (nonlinear observer‐based estimator—NLOBE, asymptotic observer with second‐order dynamics tuning—AO‐SODE) and a linear estimator (recursive least squares with variable forgetting factor—RLS‐VFF) is presented. Heterologous protein production in Pichia pastoris PAOX1 (Mut+) and PFLD1‐based systems is closely related to μ and has been addressed due to its high relevance in modern biotechnology and bioprocess engineering. μ was estimated by online gas analyses or substrate measurements, biomass, and substrate considering yield coefficients and mass balances. In simulation studies, NLOBE showed high sensitivity to tuning and initialization variables. Validation experiments demonstrated AO‐SODE performs better than the RLS‐VFF for moderate to rapid changes of μ and model parameters being known. If low changes on μ are presented, for instance, in substrate regulation, RLS‐VFF comes up as the best option, because of its reduced requirements. © 2011 American Institute of Chemical Engineers AIChE J, 58: 2966–2979, 2012