Over the past decade the interest for bioprocess monitoring using non-invasive IR spectroscopic sensors has increased considerably. This is mainly due to their rapid simultaneous multi-analyte detn. ability, in-situ sterilizability, and low need for maintenance during operation. Most of the components in culture media absorb in the MID-IR region. The complex spectra obtained can be deconvoluted using numerical methods such as Partial Least-Squares. In order to do this, a calibration model for each component needs to be generated. This model is based on an exptl. obtained calibration matrix obtained by taking the spectra of mixts. contg. different concns. of the pure components involved in the reaction. According to ASTM guidelines [1], the no. of different mixts. used in such calibration models must be at least 6 times the no. of absorbing components. This is usually very time-consuming for bioprocesses, since the no. of spectra that need to be collected is very large. One of the main advantages of using Mid-IR spectra compared to NIR is in the linear features the first present. According to Beer's law, it is possible to scale the spectrum of a pure component to obtain its spectra for different concns. Furthermore, the spectrum of a mixt. of components can be theor. achieved by adding up the spectrum of the different pure components involved. It is thus interesting to study if a calibration matrix such as the one proposed by the ASTM guidelines can be generated artificially. In this work the predictions obtained using two different calibration matrixes will be studied. The first calibration matrix contains spectra of different mixts. collected manually, while in the second matrix the spectra were generated artificially from the different pure components present in the reaction system. The two models were used to monitor a fed-batch fermn. of G. xylinus. The results obtained as well as a discussion on the role of noise will be presented. [1] Std. practices for IR quant. anal. (e 1655-97). Annual book of ASTM stds. 1999. [on SciFinder (R)]