Using case studies from climate science, this dissertation provides an analysis of the value of agreement across models, also known as "robustness." I argue for four main theses. First, that the use of multiple models is essential to the epistemology of climate science; agreement across such models is an interesting special case, but not the whole story. Second, that agreement across models and agreement across experiments should be given the same analysis. Both are fallible means of raising the probability of a hypothesis. Third, that we can understand robustness in terms of agreement across sources of evidence that vary with regards to the conditions under which they are reliable--and that, in general, such agreement will tend to confirm the agreed-on result. Fourth, and finally, that in cases where there isn't agreement across models, climate scientists can be understood as using statistical methods to determine what conclusions are justified. Such methods deserve more attention, given that--as I demonstrate--they have hidden or unappreciated presuppositions. The overarching picture is one in which models are a kind of epistemic tool, and our epistemic evaluation of them must take account of the role or purpose to which they are put.