This paper presents a scareware detection method that is based on performing data mining on extracted variable length opcode sequences derived from instruction sequences of binary files. Our experimental results show that many common supervised learning algorithms generate accurate models from subsets of our data set.