We define and study a particular kind of generalization strategy for deriving efficient functional programs. It is called higher order generalization because it consists in generalizing variables or expressions into functions. That strategy allows us to derive efficient one-pass algorithms with low time×space complexity.