The paper is dealing with models for prediction of students eficiency with the help of neuron networks and decision-making classifcation trees and then with the analysis of factors that infuence the efciency of students. A created model, based on demographic data of students as well as their behaviour and attitudes toward learning, tries to classify student in one of the two efciency categories. Te efciency is measured by the average of marks during studies. Various architectures of neuron networks have been trained and tested and the best model is obtained with the help of stratifed perceptron network. Te trees of decisi- on-making ofered a signifcantly better accuracy than neuron networks and we suggest their using due to their being a more precise method for the set of observed data. A sensitivity analysis of output variables on the input ones carried out with neuron networks refers to the fact that preliminary exams, attendance of exercises, importance of marks to students, and scholarships are among the most signifcant factors for the efciency of students. Te trees of decision-making separated the most signifcant variables: the time spent in learning, attendance of exercises and the sorts of materials from which students learn. Future researches, with the increased number of input variables and enlargement of the pattern and methodological expansion of other artifcial intelligence techniques and statistical methods, would make possible to create more succe- ssful model to be the basis for building the support system of decision-making in university level education.