In this work is presented a practical (i.e. nonlocal) robust estimation design to infer the main safety, production, and quality variables of batch, semibatch, or continuous free-radical homopolymer reactors with on-line measurements of free monomer, temperature, and volume. The solvability of the problem is established a priori, and the design features a robust nonlocal uniform convergence criterion coupled with a systematic construction-tuning procedure. The nonlocal stability framework exhibits the performance-robustness tradeoff, as well as the relationship between the sizes of the estimate error, initial state, exogenous input, and model parameter errors. The tuning of gains of the estimator can be executed with conventional filtering techniques for linear single-output filters. The polymerization of methyl methacrylate with AIBN initiator is considered as an application example, and the robust nonlocal functioning of the estimator is corroborated with numerical simulations. (C) 2000 Elsevier Science Ltd. All rights reserved.