A procedure for the extraction and capillary zone electrophoresis (CZE) separation of proteins from carob, guar and tara gums in a background electrolyte (BGE) of pH 9 containing 0.1% polyvinyl alcohol is described. The CZE protein profiles exhibit characteristic peaks for each one of the Leguminosae gums, which can be used to construct models capable of identifying samples of carob, guar and tara gums, and predicting the guar content in binary carob-guar mixtures of different geographical origin and harvested in different years. The classification and prediction models are constructed by using linear discriminant analysis (LDA) and multiple linear regression (MLR), respectively. An excellent resolution between the three categories is obtained with LDA, the model being capable of classifying samples with recognition and prediction capabilities of 100%. For MLR models constructed with carob-guar mixtures with land without a common history, the average of the calibration residuals are +/-0.50 and +/-0.90%, respectively (average values for the 2-20% guar range). For the later model, the detection limit was 3.2% guar (from the standard deviation of 18 mixtures with 2-4% guar, and for alpha = beta = 0.05).