Methods of neuroimaging have sporadically, though in recent years increasingly, occurred in legal proceedings. By now, however, it seems that they are about to enter courtrooms on a systematic basis. This poses a host of normative problems, to do, for instance, with future applications of neuroimaging to determine culpability, to test the veracity of testimony, or to predict the future dangerousness of perpetrators. The latter two, brain-based lie detection and “neuroprediction” of dangerousness, are examined in this chapter. Functional magnetic resonance imaging (fMRI) is taken as a paradigm model, and its potential impacts on criminal trials are explored. The analysis is premised on a range of basic distinctions: between (1) different phases of a criminal trial; (2) the divergent roles played by the parties to a trial, most notably prosecution and counsel, and the different evidentiary goals and burdens associated with these roles; and (3) between compulsory and consensual fMRI. It turns out that there are no good reasons to ban fMRI for lie detection or for neuroprediction from criminal proceedings entirely. Instead, it should be admitted differentially in criminal trials, viz., only for purposes of exoneration, but not of conviction, of the defendant. Substantiating arguments are expounded. In cases of preventive detention, it may even be obligatory for the state to offer chances of possibly exonerating brain imaging to perpetrators who were otherwise considered candidates for indefinite custody.