This paper presents an identification test monitoring procedure for multivariable systems whose purpose is to define an experiment that is both sufficiently informative for identification purposes and of the shortest duration possible, given predefined levels of accuracy in the model. The procedure relies on uncertainty regions resulting from frequency-domain transfer function estimation that is performed during experimental execution. To obtain independent-in-frequency signals for estimation, input design relying on multi-sinusoidal signals with “zippered” power spectra is developed. Given the various approaches available for computing statistically-based uncertainty in the frequency domain, the method that offers the most general conditions with the least a priori information about the output noise structure is selected. Based on the computed uncertainties and user-defined bounds, a stopping criterion for the identification test is developed. Results are evaluated with a simulation study involving a representative process model. This includes a performance evaluation of the technique under various distinct noise models.