We describe a lightweight behavioral malware detection technique that leverages Microsoft Windows prefetch files. We demonstrate that our malware detection achieves a high detection rate with a low false-positive rate of 1 × 10^-3, and scales linearly for training samples. We demonstrate the generalization of our malware detection on two different Windows platforms with a different set of applications. We study the loss in performance of our malware detection in case of concept drift and its ability to adapt. Finally, we measure our malware detection against evasive malware and present an effective auxiliary defensive technique against such attacks.