We use Find-S or Least general algorithm to find initial generalized association rules, then use it to generate the initial population. With the two operators of crossover and mutation, we run the genetic algorithm to generate new association rules. The problem is that genetic algorithm is randomize algorithm. So we need to scan the database many passes in order to make it converge. That is expensive. But if we can find a good initial population, then the number of the passes to scan the database may be not many.