What makes software fail? This talk describes techniques for creating effective malicious inputs designed to break network protocol and file format implementations. We look at the categories of destructive data, map them to practical examples taken from common network protocols and file formats, and analyze past vulnerabilities that could be triggered by these inputs. For selected vulnerabilities, we look at source code to understand how the malicious inputs reach the problematic portions of code and how each vulnerability is ultimately triggered. The analyzed interfaces include both wired and wireless protocols as well as various file formats including images, video and compression formats and X.509 certificates. We observe that only a few relatively simple insights, combined with a solid knowledge of common categories of data structures, already helps break most implementations through rigorous negative black-box testing. From a proactive perspective, understanding what kinds of inputs typically trigger common programming errors also helps developers protect against attacks and to avoid most common problems.