This thesis is concerned with the classification of protein motifs based on their secondary structure. A protein motif is a short sequence of amino acids that has remained partially conserved throughout evolution and can be associated with a recognizable part of the protein structure performing a distinct function. The goal is to improve the accuracy of motif scanning procedures by detecting as many significant motifs as possible (true positives) while minimizing the number of wrong assignments (false positives). To enhance accuracy, we employ an approach based on pairwise similarity, examining not just the similarity to the query but also the mutual similarity among protein motifs. 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 protein motifs sharing the same secondary structure annotation will group together and be located within a sphere. By finding this sphere, we obtain points such that each pair is “sufficiently” close to one another. However, we encounter two key challenges: the annotation for protein secondary structure is not reliable, and we lack a characteristic query. We addressed these challenges to the best of our ability, but due to unreliable annotation, we couldn't fully test the method. We showed that true positives do, in fact, cluster and could be found using this method. However, for this approach to be fully effective, we need more accurate annotation and a better characteristic query.