We have reported the classification of infrared spectra of binary mixtures of 4-n-nonyl-4’-cyanobiphenyl (9CB) and 4-n-pentylphenyl-trans-4’-pentylcyclohexane-1-carboxylate (5H5) using back propagation networks (BPN)(1). For the analysis of large sets of data, such as in the cases of infrared spectra of mixtures of 9CB and 5H5, the Gaussian radial basis function networks (RBFN) are better than the BPN. The input neurons represent selected spectral intensities in the range of 1000-40000 wave numbers/cm. The output neurons represent the concentrations of the mixtures, the phases (isotropic, nematic and smectic) at the experimental temperatures. The RBFN correctly classify the phases. The testing for the compositions of some mixtures show improvement over the back propagation networks.