Adults in the United States have a low level of science literacy, and public consensus on major scientific issues like climate change and evolution is hampered by pervasive misinformation and “fake” science on the Internet, often spread by social media. The situation represents a threat to the functioning of civic society. The paper reports on a project to combat scientific misinformation by automatically identifying it using machine learning. Neural networks were trained using sets of non-technical articles selected by science undergraduates on the Internet, with equal numbers of articles containing legitimate science and science misinformation. Climate change and evolution were used as topics for this testbed. After experimenting with various machine learning algorithms, an accuracy above 90% was achieved for the neural net identifying the real science content. In the next phase of the project, this technology will be scaled to large samples of content drawn from CommonCrawl, and it will be applied across more domains of science. Then it will be deployed as a web browser extension that presents the probability that a particular web page has real or fake science, and as a smartphone app for that allows users to classify articles as real or fake.