This paper describes the ideas and methods that led to the writing of analgorithmic composition NTrope Suite. This piece, for solorecorders and voice, was generated by ‘mixing’ works bydifferent composers from different eras. The idea behind the work was toexamine random generation procedures that could maintain stylisticproperties typical to the reference works. The interesting property ofthis method is that it implements a sort of ‘statisticallearning’, that optimally preserves the properties of the referencepieces and also properly ‘generalises’ them so as to create anew valid work. The musical result is very coherent, maintaining bothstylistic resemblance to the reference music and exhibiting some surprisingoriginality as well. Theoretically, the resulting piece is closest to thereference works in terms of mutual entropy. The algorithm and itstheoretical significance are discussed in the paper.