A methodology is proposed for selecting parameters to estimate when data are too limited to estimate all kinetic,thermodynamic, and mass transfer parameters. in complex models Of chemical processes. When data are sparse, noisy, or correlated, it is often letter to obtain predictions from a simplified model (SM) where a few parameters have been removed via simplifying assumptions or some parameters are fixed at nominal values based on prior knowledge. Reducing the number of estimated parameters leads to bias in model predictions, but also lowers prediction variance. Trade-off between bias and variance is assessed using the mean squared error (MSE) of the model predictions. The proposed model selection criterion is an advance over previous criteria in the literature because arbitrary tuning parameters are not required, computations are relatively simple, and the user can specify key operating conditions where accurate predictions are desired. Important benefits are that overfitting of noisy data is prevented and standard least-squares parameter estimation can be used without numerical difficulties. Monte Carlo simulations are used to assess the effectiveness of the proposed methodology for parameter selection in linear and nonlinear models. This approach will be valuable for industrial modelers who want to make accurate predictions about new product specifications or grades.