The main cause of several neurodegenerative diseases such as Alzhemier, Parkinson and spongiform encephalopathies is the formation of amyloid fibrils in proteins. The systematic analysis of amyloid and non-amyloid sequences provide deep insights on the preference of amino acid residues at different locations of amyloid and non-amyloid peptides. In this work, we have systematically analyzed 139 amyloid and 168 non-amyloid hexapeptides and revealed the preference of residues at six different positions. We observed that Glu, Ile, Ser, Thr and Val are dominant at positions six, five, one, two and three, respectively in amyloid peptides. In non-amyloids, similar trend is noticed for few residues whereas the residues Ala in position 2, Asn in position 4, Gly in position 6 etc, showed different trends to that of amyloids. Utilizing the positional preference of 20 amino acid residues, we devised a statistical method for discriminating amlyloids and non-amyloids, which showed an accuracy of 89% and 54%, respectively. Further, we have examined various machine learning techniques, and a method based on Random Forest showed an overall accuracy of 99.7% and 83% using self-consistent and 10-fold cross-validation, respectively using the positional preference of amyloids and non-amyloids along with three selected amino acid properties. © 2012 Springer-Verlag.