A variety of physically based, statistical, and machine learning techniques have been used to retrieve atmospheric temperature profiles from meteorological satellite sounder data. This paper uses a boosted regression tree for evaluating a function that determines the temperature at each level from the sounder data. For large data sets it is demonstrated that this is more accurate than the conventional approaches. For a large data set it is slightly more accurate than the most accurate feed-forward neural network strategy, using the same data. With a neural network the separation of the data for use with different networks in different regions in parameter space is an advantage. In contrast a boosted regression tree has slighly smaller errors for a single model. This occurs as an increase in the size of the data set tends to increase the accuracy. The relative advantages and disadvantages of feed-forward neural networks and regression trees are discussed.