Аннотация:Algorithm which has taken the third place in “JRS 2012 Data Mining Competition” among 126 participants is described. The competition was related to the problem of predicting topical classification of scientific publications in a field of biomedicine. The presented algorithm is a combination (blend) of simple classification algorithms: a linear classifier, a k-NN classifier and two SVMs. We build the combination using special estimation matrices. It proves again that combinations have significantly better performance compared to their individual members.