The use of Fourier transform mid-infrared spectroscopy (FT-MIRS) to predict the concentrations of key analytes in fed-batch cultivations of an industrial strain of Pichia pastoris in a chemical complex medium was investigated. Models for glycerol, methanol (substrates), and product (an heterologous protein) were built, and evaluated. The use of a multi-bounce attenuated total reflectance (HATR) accessory aided spectral acquisition in optically dense samples. Generally, all models were robust and performed well on external validation, using data from processes not present in the original modelling exercise. Substrate models lacked the complexity of some previous IR models, and the models performed adequately even at low analyte concentration (<1 g l(-1)). Thus, simultaneous, rapid monitoring of low concentrations of multiple analytes in a complex bioprocess matrix with little or no sample pre-treatment is achievable using ATR FT-MIRS.