A single spectra library was used to monitor on-line, by mid-infrared spectroscopy, nine different batch cultures of Escherichia coli performed with various medium compositions, including chemically complex formulations. Whereas the classic chemometrics approach would have required the preparation and measurement of hundreds of standards, only six spectra were included in the library. These included the molar absorbance of the four main metabolites (i.e. glucose, glycerol, ammonium and acetate), and the remaining two were drift spectra found by factor analysis. The accuracy of the prediction was not altered by a change of the carbon source, the ammonium concentration or even the addition of chemically undefined compounds, such as yeast extract and peptone. The standard errors of prediction. averaged over the nine experiments were 8.0, 12.3, 5.9 and 5.6 mM for glucose, glycerol, ammonium and acetate, respectively. Inclusion of two drift spectra in the library provided an estimation of how noisy an experiment was. This also allowed detection of batch cultures that require further investigation, namely runs which were subject to large signal drift or during which an unexpected compound was produced, without having to carry out time-consuming off-line analyses. (C) 2008 Elsevier B.V. All rights reserved.