In this paper, the variable structure model (VSM) approach is applied to modeling and parameter estimation in processes exhibiting simple and complex relaxation oscillations. The approach reduces the full model to two simple submodels and permanently chooses the submodel which has a better performance in the description of the input-output behavior of the process. The degree to which the approach will adequately approximate the true process dynamics increases with the system stiffness. Excellent performance of the approach is demonstrated in both online and off-line parameter estimation for a single CSTR as well as for a train of reactors with the Belousov-Zhabotinski or polynomial kinetics.