In 1994, Eli Biham and Alex Biryukov made significant improvements to the technique.
It is a known-plaintext attack based on the non-uniform distribution of the outputs of pairs of adjacent S-boxes.
It works by collecting many known plaintext/ciphertext pairs and calculating the empirical distribution of certain characteristics.
There are tradeoffs between the number of required plaintexts, the number of key bits found, and the probability of success; the attack can find 24 bits of the key with 252 known plaintexts and 53% success rate.
In 1998, Pornin developed techniques for analyzing and maximizing a cipher's resistance to this kind of cryptanalysis.