Partitioning cryptanalysis
A generalisation of linear cryptanalysis that sorts plaintexts and ciphertexts into classes and measures how unevenly they line up.
Carlo Harpes and James Massey proposed the method in 1997. In linear cryptanalysis the "class" is always the parity of some chosen subset of bits — two buckets, and you hope the result lands in one of them slightly more often. Partitioning cryptanalysis allows arbitrary splits into more buckets, giving more freedom to find whatever imbalance a particular cipher happens to have.
That freedom is both strength and weakness: the space of possible partitions is enormous, and there is no general way to find the best ones. The technique is therefore used mostly to argue resistance against the whole family of linear-style attacks, and to explain why special cases such as mod n cryptanalysis work at all.