Opcodes as predictor for malware

This paper discusses a detection mechanism for malicious code through statistical analysis of opcode distributions. A total of 67 malware executables were sampled statically disassembled and their statistical opcode frequency distribution compared with the aggregate statistics of 20 non-malicious samples. We find that malware opcode distributions differ statistically significantly from non-malicious software. Furthermore, rare opcodes seem to be a stronger predictor, explaining 12-63% of frequency variation.