We show that state-of-the-art self-supervised language models can be readily\nused to extract relations from a corpus without the need to train a fine-tuned\nextractive head. We introduce RE-Flex, a simple framework that performs\nconstrained cloze completion over pretrained language models to perform\nunsupervised relation extraction. RE-Flex uses contextual matching to ensure\nthat language model predictions matches supporting evidence from the input\ncorpus that is relevant to a target relation. We perform an extensive\nexperimental study over multiple relation extraction benchmarks and demonstrate\nthat RE-Flex outperforms competing unsupervised relation extraction methods\nbased on pretrained language models by up to 27.8 $F_1$ points compared to the\nnext-best method. Our results show that constrained inference queries against a\nlanguage model can enable accurate unsupervised relation extraction.\n