In the present work, the contents of 38 elements of 65 vitex (Vitex negundo var. heterophylla Rehd.) honey samples from Shunyi of Beijing, Fuping and Pingshan of Hebei province were determined by inductively coupled plasma mass spectrometry (ICP-MS). Among them, B, Na, Mg, P, K, Ca, Fe and Zn were the most abundant elements with mean contents more than 1 mg kg-1. It can be found that there were relationships between the contents of elements and the geographical origin of vitex honey samples. Taking the contents of 29 out of 38 mineral elements (Na, Mg, Al, K, Ti, V, Mn, Fe, Ni, Cu, Zn, Ga, As, Sr, Y, Mo, Cd, Ba, La, Ce, Pr, Nd, Sm, Gd, Dy, Ho, Tl, Pb and U) as variables, the chemometric methods, such as principal component analysis (PCA) and back-propagation artificial neural network (BP-ANN), were applied to classify vitex honey samples according to their geographical origins. PCA reduced all of the variables to four principal components and could explain 81.6% of the total variances. The results indicated that PCA could mainly classify the vitex honey samples into three groups. BP-ANN was explored to construct classification model of vitex honeys according to their geographical origin. For the whole data set, the overall correct classification rate and cross-validation (leave one out method) rate of proposed BP-ANN model was 100% and 95.4%, respectively. To further test the stability of the model developed for prediction, 75% of honey samples of each geographical origin were randomly selected for the model training set, and the remaining samples were classified with the use of the constructed model. Both the overall correct classification rate and prediction rate of proposed BP-ANN model were 100%. It is concluded that the profiles of multi-element by ICP-MS with chemometric methods could be a potential and powerful tool for the classification of vitex honey samples from different geographical origins.