The results of the classifier optimization based on the Karhunen-Loeve transform (KLT) 
are shown. As optimization parameters the spectral components vector of the KLT and 
the geometrical Minkovski dissimilarity measure were taken. The optimization result 
is the average prototypes number minimizing that is needed for correct test pattern 
identifying with given probability value. The prototypes number is determined basing 
on the sequential analysis procedure.