A method of de-convoluting a mixture composition from a set of property data for the mixture is presented and evaluated. This is the probability apportioning method (PAM version 4). It is compared with the well-known spectral analysis method of least squares (LSQ). Both these methods assume that the component data is combined in the mixture in a linearly independent manner. PAM starts with an assumed composition and uses Bayes' theorem on conditional probabilities in an iterative procedure to recalculate the prediction. LSQ is shown to be generally more applicable, but PAM is less susceptible to noise where there are components in the mixture with similar property data (spectra).