Regression trees are a type of decision tree used for predicting continuous values by recursively partitioning the data space into regions with similar output values. They are particularly useful for capturing non-linear relationships and interactions between variables without requiring extensive data preprocessing.