To determine if a prediction of epidemic cholera using climate data can be made, we performed autoregression analysis using the data recorded in Dhaka City, Bangladesh over a 20-year period (1983-2002) comparing the number of children aged <10 years who were infected with Vibrio cholerae O1 to the maximum and minimum temperatures and rainfall. We formulated a simple autoregression model that predicts the monthly number of patients using earlier climate variables. The monthly number of patients predicted by this model agreed well with the actual monthly number of patients where the Pearson's correlation coefficient was 0·95. Arbitrarily defined, 39·4% of the predicted numbers during the study period were within 0·8-1·2 times the observed numbers. This prediction model uses the climate data recorded 2-4 months before. Therefore, our approach may be a good basis for establishing a practical early warning system for epidemic cholera. © 2007 Cambridge University Press.