I examine the causes and consequences of environmental beliefs. Rice farmers in Bangladesh must learn about their soil’s salinity to choose the appropriate seed. Comparing beliefs to agronomic readings, I document both significant over- and underestimation of salt levels. A simple identification problem explains this pattern: farmers learn about multiple unobserved threats from ambiguous signals. Bayesian farmers endogenously process data in support of their priors, e.g., someone worried about high salinity will interpret low yield as a sign of too much salt. Climate change amplifies this process by systematically altering the risks considered most threatening. I confirm the framework’s predictions using a lab-in-the-field exercise and natural experiments that isolate salient shocks that capture attention (e.g., tidal flooding) and subtle shifts that go unnoticed (e.g., irrigation water contamination through rising sea-levels). Despite equal effects on true salt levels, salient saltwater floods increase salinity beliefs substantially more than does subtle irrigation intrusion. These experiences shape how farmers interpret new data: past exposure to salient shocks increases the mental link between low yield and salinity while subtle shocks reduce the perceived diagnosticity of salinity clues. In large-scale field experiments, correcting misperceptions significantly alters farmers’ demand for salinity-tolerant seeds with substantial consequences for profits.