Abstract

We show that Corollary 1 of “On Multiple Linear Approximations” (Crypto 2004 – LNCS 3152) is incorrect.  
In particular, the value given for the gain by Corollary 1 is likely to be a significant overestimate of this quantity.  
Thus any data requirements for linear cryptanalysis with multiple linear approximations based on this value for the gain are highly questionable.