Transient diffuse reflectance spectra in the visible wavelength region from an enzyme optode were evaluated to determine artificial mixtures consisting of penicillin and ampicillin simultaneously. In order to achieve a discrimination of the samples the optode operated at two different working pH values. Using an appropriate FIA modus a linear range in the increase and decrease of the time-dependent signal was registered which could be easily fitted by means of ordinary linear regression. In this way characteristic static and dynamic features of the enzyme catalyzed reaction could be extracted for the evaluation by partial least-squares (PLS). The pre-processed raw data were differently weighted and the optimal number of primary factors was each selected by full cross validation. The interpretation of the spectral loadings and the correlation matrix identifies redundant variables resulting in a model with an optimal number of variables. The results obtained with this model show that the straightforward linear regression can be a sufficient approach to avoid the introduction of systematic errors in on-line pre-processing of data.