Over the past 20 years, predicates of personal taste (PPT) have been at the centre of a set of lively debates in the philosophy of language and linguistics. These debates have yielded many subtle and inventive analyses of PPT. There is, however, a crucial methodological question about PPT that remains underexplored: what sorts of evidence should be used in evaluating an account of PPT? In line with a fairly large body of recent work, we point out that since hypotheses about PPT are empirical, they need to be evaluated empirically, using clearly articulated tests. The test on which we focus involves ordinary speakers’ judgments about retraction. We identify what we take to be two significant problems with the leading account, due to John MacFarlane, of why retraction data matter in the PPT debates. We then develop an improved framework for thinking about this issue. To close, we use our improved framework to re-evaluate the findings on retraction reported by Markus Kneer, and we discuss a few loose ends that will need to be addressed by future experimental research on PPT.