We present ELEM2, a new method for inducing classification rules from a set of examples. The method employs several new strategies in the induction and classification processes to improve the predictive performance of induced rules. In particular, a new heuristic function for evaluating attribute-value pairs is proposed. The function is defined to reflect the degree of relevance of an attribute-value pair to a target concept and leads to selection of the most relevant pairs for formulating rules. Another feature of ELEM2 is that it handles inconsistent training data by defining an unlearnable region of a concept based on the probability distribution of that concept in the training data. To further deal with imperfect data, ELEM2 makes use of the post-pruning technique to remove unreliable portions of a generated rule. A new rule quality measure is proposed for the purpose of post-pruning. The measure is defined according to the relative distribution of a rule with respect to positive and negative examples. To show whether ELEM2 achieves its objective, we report experimental results which compare ELEM2 with C4.5 and CN2 on a number of datasets.