We analyzed 50 yr of cholera mortality data from 26 districts in historic Bengal using mechanistic models and recently-developed likelihood inference techniques. Because our statistical methodology makes few assumptions about the form of the model, we are able to incorporate heretofore unconsidered effects including an environmental pathogen reservoir and different modes of immunity. Our models fit the data dramatically better than all previously presented models and robustly predict that most exposures result not in infection but in short-term immunity, which wanes on a timescale of weeks to months. We quantify the relative importance of human-human vs. environmental transmission and evaluate the role of multiannual climate drivers in cholera transmission.