The outputs of four models (Euler numerical model, neural network model, dynamic statistical model, consensus forecast model) for forecasting precipitation pH from May to September 2005 are analyzed and their effectiveness is compared. The analysis concludes that the Euler numerical model often produces data overflow; the four models all has problems in localization; the forecasting accuracy is affected by the local basic precipitation pH level; the formation mechanisms of acid rains in the coast and inland cities differ from each other; in some cities, some models without considering ions performed as well as the models with ions.