Major League Baseball of the USA is considered the most competitive and challenging arena of baseball. And the population of baseball fans is still increasing. Many scholars and fans are interested in using each team’s performance data to predict outcomes of MLB games. Their prediction accuracy is around 50%. Our goal is to use performance data to predict the yearly winning percentage of each team. Our research method is Classification and Regression Trees (CART) and Maximum Likelihood Regression Trees (MLRT). In addition, we will discuss the prediction accuracy of the CART and MLRT models, and apply the result to predict the playoffs list of the MLB. We find that these models all have good prediction effectiveness for the yearly winning percentage with the MAPE between 12% and 13%. CART models are slightly better than MLRT models in winning percentage prediction. For the playoffs prediction, MLRT is only better than CART in 2018 for models eliminating collinear variables. The prediction effectiveness of CART is the same or better than the MLRT for the rest.