A trademark confusion survey is a measuring instrument, not an opinion, and its number is only as trustworthy as the design choices behind it. This guide explains, in plain language, how confusion surveys are engineered and how to read them the way a cross-examiner does—starting with the survey universe (the choice that most often decides the case), then the Eveready, Squirt, and aided-Eveready formats, the indispensable control group and the net-confusion figure it produces, marketplace-realistic stimuli, non-leading question design, double-blind administration, and probability versus non-probability sampling. It maps how courts test survey methodology under Federal Rules of Evidence 702, 703, and 403 and the Daubert trilogy after the 2023 amendment, what net-confusion levels courts actually credit, and the specialized formats for secondary meaning, fame, and genericness. Worked hypotheticals and a deep bench of real decisions—from Cracker Barrel and Borghese to 1-800 Contacts and J.T. Colby v. Apple—show how a single design choice can flip the result. It closes with a reading checklist, an FAQ, and separate playbooks for the proponent and the challenger.