Combining information from various sources is a requirement for solving many real-world problems, and it is of interest to study the conditions under which improved performance ispossible. In this talk labelling techniques for reducing correlation in the context of multiple classifier systems will be discussed. First an existing technique, ECOC (Error Correcting OutputCoding) will be explained. ECOC is an information theoretic concept that uses binary code words to decompose a multi-class problem into a set of complementary two-class problems. Then three novel labelling techniques will be described. The first is a variant of ECOC that reduces sensitivity to code word selection. The second technique partitions the data by assigning a binary label to each training pattern, and then learning a set of combining weights by treating the binary labels as target values. The third technique maps each training pattern to a vertex of the binary hypercube, enabling identification of a set of inconsistently classified patterns, which are used to randomly perturb classifier training sets. Experimental evidence on real and artificial benchmark data with RBF and MLP as base classifiers, demonstrates improvement due to these three techniques.