The size of databases seem to be ever increasing. Most machine learning algorithms have been created for handling only a small training set, for instance a few hundred examples. In order to use similar techniques in databases thousands of times bigger, much care must be taken. Having very much data is advantageous since they probably will show relations really existing, but the number of possible descriptions of such a dataset is enormous. Some possible ways of coping with this problem, are to design algorithms with lower complexity and to use heuristics to find the best classification rules. Simply using a faster computer is seldom a good solution.