Disagreements about substance and expert credibility often go hand-in-hand and are hard to resolve, even when people share common information, on a wide range of issues ranging from economics, climate change, to medicine. We argue that a learning bias helps explain disagreement in environments such as these where both the state of the world and the credibility of information sources (experts) are uncertain. Individuals with our learning bias overinterpret how much they can learn about two sources of uncertainty from one signal, leading them to over-infer expert quality. People who encounter information or experts in different order disagree about substance because they endogenously disagree about the credibility of each others' experts. Disagreement persists because first impressions about experts have long-lived influences on beliefs about the state. These effects arise even though agents share common priors, information, and biases, providing a theory for the origins of disagreement.