This thesis studies the protein family classification problem. Proteins are classified by identifying a short amino acids sequence contained within them. This short sequence is called a motif. During evolution, the motif has remained partly conserved and can be associated with the recognizable part of the protein structure that performs a distinct function. Unlike typical motif scanners that observe the similarity of potential motifs with a given motif, this thesis describes and tests the method that considers pairwise similarity. The method works in combination with another motif scanner by providing additional criteria for motif identification. By describing amino acids using numerical vectors the problem is placed in Euclidean space where distance is considered instead of similarity. The assumption is that the motifs, found in proteins that belong biologically to the observed family, will group. The goal of this method is to find the optimal ball that will contain as many correct hits and as few wrong ones as possible. The method was tested on twelve plant proteomes whose proteins were searched for containing characteristic motifs of the VQ protein family. In most of the cases, the method proved to be successful which in turn confirmed true positives clustering assumption.