Civil commitment, confinement under sexual predator laws, and many capital and noncapital sentences depend upon proof of a propensity toward violence. This article discusses the current state of prediction science, in particular the advantages and disadvantages of clinical and actuarial prediction, and then analyzes how the rules of evidence should be interpreted in deciding whether opinions about propensity should be admissible. It concludes that dangerousness predictions that are not based on empirically-derived probability estimates should be excluded from the courtroom unless the defense decides otherwise. This conclusion is not bottomed on the usual concern courts and commentators raise about expert prediction testimony–that the associated false positive rates are too high. In fact, because they produce better-than-chance results, both clinical and actuarial risk assessments are, with certain caveats, sufficiently probative to meet the expertise threshold, whether that threshold is defined by Frye or Daubert. Rather, the rationale for the conclusion reached in this article is that prediction testimony should only be inadmissible when its prejudicial impact outweighs its (admittedly weak) probative value. Because of the strong predilection in favor of a dangerousness finding in commitment and sentencing proceedings, that formula means that the defense should be able to use either type of prediction testimony in its case-in-chief, while the government should only be able to use actuarial testimony, which empirical evidence suggests is more likely to be taken for what it is worth even when it bolsters the state’s case.