In this paper we propose a method of association rule mining using Genetic Network Programming (GNP) with a self-adaptation mechanism in order to improve the performance of association rule extraction systems. GNP is a kind of evolutionary methods, whose directed graphs are evolved to find a solution as individuals. Self-adaptation behavior in GNP is related to adjust the setting of control parameters such as crossover and mutation rates. It is called self-adaptive because the algorithm controls the setting of these parameters itself - embedding them into an individual's genome and evolving them. The aim is not only to find suitable adjustments but to do this efficiently. Our method can measure the significance of the association via the chi-squared test and obtain a sufficient number of important association rules. Extracted association rules are stored in a pool all together through generations and reflected in three genetic operators as acquired information. Further, our method can contain negation of attributes in association rules and suit association rule mining from dense databases.