Algal blooms have caused severe problems in Lake Taihu, China. Early warning of phytoplankton accumulation can support decision-making against harmful algal bloom events. To investigate the performance of different models in forecasting high phytoplankton biomass, we developed a mechanistic, a regression and three artificial neural network (ANN) models to predict short-term (3 days) changes of phytoplankton biomass (expressed as chlorophyll-a concentration) in Gonghu Bay of Lake Taihu. We determined the input variables of the ANN models with a sensitivity analysis, and optimized their parameters with a trial-and-error approach. The sensitivity analysis revealed the effects of the input variables on phytoplankton biomass. To calibrate and validate the models, we collected two data sets of Lake Taihu in 2009: hourly-averaged data collected by an automatic monitoring system and field data with a sampling interval of twice a week. Although the sensitivity analysis results vary among the five models, there is a general consensus that phytoplankton changes are significantly affected by water temperature in Gonghu Bay. The ANN models obtained good model fit indicating their practical values in predicting non-linear phytoplankton dynamics for water management purpose. The mechanistic model predicted the phytoplanIcton distribution dynamically and described the variable interactions explicitly. The regression model is characterized by its easy development. This comparison study assists the modelers in selecting an approximate model for their specific purposes.