BART is a Bayesian nonparametric regression method that uses a sum of regression trees to model the unknown function f(x) = E(Y|x). Each tree is constrained by a regularization prior to be a weak learner, and the model is fitted using an iterative Bayesian backfitting MCMC algorithm. This approach allows for full posterior inference, including point and interval estimates of the regression function and marginal effects of predictors. BART can also be used for model-free variable selection by tracking predictor inclusion frequencies. The method is illustrated with a bake-off against competing methods on 42 data sets, a simulation experiment, and a drug discovery classification problem. The BART model consists of a sum-of-trees model combined with a regularization prior. A Bayesian backfitting MCMC algorithm is used for inference, and a probit extension is provided for classification. The model is implemented in R as the BayesTree library. BART outperforms other methods in predictive performance across various data sets and is robust to hyperparameter settings. The method is effective in detecting low-dimensional structures in high-dimensional data and provides accurate estimates of partial dependence functions and variable importance.BART is a Bayesian nonparametric regression method that uses a sum of regression trees to model the unknown function f(x) = E(Y|x). Each tree is constrained by a regularization prior to be a weak learner, and the model is fitted using an iterative Bayesian backfitting MCMC algorithm. This approach allows for full posterior inference, including point and interval estimates of the regression function and marginal effects of predictors. BART can also be used for model-free variable selection by tracking predictor inclusion frequencies. The method is illustrated with a bake-off against competing methods on 42 data sets, a simulation experiment, and a drug discovery classification problem. The BART model consists of a sum-of-trees model combined with a regularization prior. A Bayesian backfitting MCMC algorithm is used for inference, and a probit extension is provided for classification. The model is implemented in R as the BayesTree library. BART outperforms other methods in predictive performance across various data sets and is robust to hyperparameter settings. The method is effective in detecting low-dimensional structures in high-dimensional data and provides accurate estimates of partial dependence functions and variable importance.