Standards published by the Association of Forensic Science Providers (2009, Standards for the formulation of evaluative forensic science expert opinion, Science & Justice, Vol. 49, pp. 161–164) encourage forensic scientists to express their conclusions in the form of a likelihood ratio (LR), in which the value of the evidence is conveyed verbally or numerically. In this article, we report two experiments (using undergraduates and Mechanical Turk recruits) designed to investigate how much decision makers change their beliefs when presented with evidence in the form of verbal or numeric LRs. In Experiment 1 (N = 494), participants read a summary of a larceny trial containing inculpatory expert testimony in which evidence strength (low, moderate, high) and presentation method (verbal, numerical) varied. In Experiment 2 (N = 411), participants read the same larceny trial, this time including either exculpatory or inculpatory expert evidence that varied in strength (low, high) and presentation method (verbal, numerical). Both studies found a reasonable degree of correspondence in observed belief change resulting from verbal and numeric formats. However, belief change was considerably smaller than Bayesian calculations would predict. In addition, participants presented with evidence weakly supporting guilt tended to “invert” the evidence, thereby counterintuitively reducing their belief in the guilt of the accused. This “weak evidence effect” was most apparent in the verbal presentation conditions of both experiments, but only when the evidence was inculpatory. These findings raise questions about the interpretability of LRs by jurors and appear to support an expectancy-based account of the weak evidence effec