The work reported illustrates an approach to the estimation of uncertainty in multivariate vibrational spectroscopic data. The definition and theory of uncertainty used is derived from the individual contributions to uncertainties. The approach adopted involves the characterization and modelling of random effects in infrared spectroscopic data. Spectral phenomena present in the data were described by models of uncertainty. These models were subsequently collated and used successfully for the predictions of sample concentrations, supplementing them with data on estimated uncertainty. Further, the selectivity of the data and the wavelength combination in terms of minimal and optimal sensors, were also calculated and shown to be influencing the uncertainty of concentration estimates. The relationship between selectivity and uncertainty was investigated and the dependence was shown to be an inverse linear function. Finally, the calibration approach employed was shown to be largely independent of the effects of smoothing.