The potential for noninvasive determinations of glucose, lactate, and ammonia in cell culture media was investigated through near-infrared spectroscopy of the 2.0–2.5 μm combination region. Samples taken from a three-day fibroblast culture were analyzed with standard clinical chemistry techniques and with a Fourier transform infrared spectrometer. The data were analyzed in three phases in an attempt to solve problems of correlation in the data. First, partial least-squares regression was used to build a multivariate calibration model based on the near-infrared spectra of the cell culture media and measured reference values. Second, an independent calibration was performed with aqueous mixtures of glucose, lactate, ammonia, glutamate, and glutamine. Finally, a calibration using a combination of spectra from cell culture media samples and aqueous mixtures was performed. In each case, the spectral loadings obtained from the calibration were inspected to determine the extent of weighting given to different spectral regions. The results indicate that a selective calibration model can be produced by combining data from samples of different type. Prediction errors of 2, 8, and 15% were obtained for glucose, lactate, and ammonia, respectively.