It’s a worthy research topic to use genetic algorithm for classification rules in data mining. In this paper, it was studied and researched in-depth. Firstly, we combined genetic algorithm and machine learning together, and then analyzed architecture of the genetic algorithm-based classification system, and also its development concrete structure was given. Secondly, we proposed a data classification rules learning system based on adaptive genetic algorithm, which can learn the classification rules accurately from the dataset. Finally the standard Play Tennis dataset was used for a closed test and after learning the system got three classification rules all with 100% accuracy rate, which fully demonstrated the feasibility of this algorithm.