DNA has long been considered the gold standard of forensic evidence, heralded for its ability to exonerate the innocent and convict the guilty. But this new generation of DNA evidence is far from its established predecessor — both in the quality of the evidence collected and the clarity of what is presented in court. With new highly sensitive technology, tiny amounts of DNA, often just a few cells, are now collected at crime scenes. This DNA was left on objects by someone who touched an object, or by someone who touched or was touched by someone who then touched the object, or even further removed. In other words, cells travel and can be easily transferred to an object without the person who was the source of that DNA ever having come into contact with the object. But this is only the beginning of the difference. These crime scene samples often contain multiple people's DNA, and there is degraded and missing information. The samples are incomplete and mixed together. In order to give some evidentiary weight to these complicated and incomplete mixtures, crime labs are turning to something called probabilistic genotyping. These computer programs generate something called a likelihood ratio. These likelihood ratios purport to express the probabilistic relationship between two hypotheses, the hypothesis that the suspect is in the DNA sample compared to the hypothesis that the suspect is not included in the sample. In this paper, I first explain what a likelihood ratio is, how it is generated, and some of the fundamental problems with the evidence. Then I turn to an analysis of the propriety of this evidence in criminal trails. It is my position that because likelihood ratios can only be generated by first presuming guilt (inclusion), they undermine the presumption of innocence, and that, by weighing these hypotheses equally, they water down the burden of proof beyond a reasonable doubt in a criminal trial. This is complicated by the sheer power of the DNA moniker and opacity of the numbers generated.