In this paper, first we extend two basic measures of association rules in data mining (i.e, confidence and support) to the case of fuzzy association rules. The main difference between standard and fuzzy association rules is the discretization of continuous variables. While continuous variables are divided into intervals for generating standard association rules, they are divided into linguistic values in the case of fuzzy association rules. Next we examine two specifications of rule weights of fuzzy association rules for pattern classification problems. One is the direct use of the confidence as a rule weight. The other is based on a slightly complicated formulation where the rule weight of each fuzzy association role is discounted by the confidence or other rules with the same antecedent conditions and different consequent classes. Through computer simulations on a pattern classification problem with many continuous attributes, we compare these two definitions with each other. Simulation results show that the direct use of the confidence is inferior to the other definition of rule weights. Then we examine three rule selection criteria (i.e., confidence, support, and their product). It is shown that good fuzzy association rules are extracted from numerical data using the product criterion. Finally we compare the performance of fuzzy association rules with that of standard association rules.