In classification problems, trees are mostly known in the form of decision trees. In the simplest case, the classifier consists of a single tree whereby each class can be described as a set of paths leading to its prediction—usually depicted as a set of conjunctive rules. An alternative usage are pattern trees, consisting of logical operators as inner nodes and input features as leaf nodes. Each such tree thus encodes a Boolean function on the input features. While pattern trees were introduced a few years ago in a fuzzy context, we propose a greedy algorithm learning pattern trees limited to alternating the Boolean operators AND and OR, which can also be transformed into compact deep rule concepts. Our experiments on UCI data sets indicate that the learned alternating Boolean pattern trees achieve similar performance as state-of-the-art rule learners. In particular on categorical data, they outperform fuzzy pattern trees in both accuracy and efficiency, which, on the other hand, have more effective operators for dealing with numeric attributes. We also demonstrate how these pattern trees may be interpreted as deep rule sets, which do not directly link input features to a prediction but do so via automatically formed intermediate concepts.