Constructing beliefs about the world usually requires simplifying assumptions. We analyze the beliefs of agents who make reasonable assumptions to model a complex situation and who make predictions conditional on those assumptions. Our theory identifies tight connections between model uncertainty (the extent to which different models lead to different predictions), overprecision (too-small variance estimates), and interpersonal disagreement (variance in mean predictions). We test these predictions in an experiment in which participants view a scatterplot and report mean and variance estimates for out-of-sample predictions. Consistent with our theory, different people focus on different plausible models but provide reasonable estimates of uncertainty conditional on their model. As a result, model uncertainty increases both overprecision and disagreement. Outside of the lab, we find similar evidence in the Survey of Professional Forecasters, including that overprecision positively covaries with disagreement.