Mechanistic modeling of how algal species produce metabolites (e.g., taste and odor compounds geosmin and 2-methyl isoborneol (2-MIB)) as a biological response is currently not well understood. However, water managers and water utilities using these reservoirs often need methods for predicting metabolite production, so that appropriate water treatment procedures can be implemented. In this research, a heuristic approach using Adaptive Network-based Fuzzy Inference System (ANFIS) was developed to determine the underlying nonlinear and uncertain quantitative relationship between observed cyanobacterial metabolites (2-MIB and geosmin), various algal species, and physical and chemical variables. The model is proposed to be used in conjunction with numerical water quality models that can predict spatial-temporal distribution of flows, velocities, water quality parameters, and algal functional groups. The coupling of the proposed metabolite model with the numerical water quality models would assist various utilities which use mechanistic water quality models to also be able to predict distribution of taste and odor metabolites, especially when monitoring of metabolites is limited. The proposed metabolite model was developed and tested for the Eagle Creek Reservoir in Indiana (USA) using observations over a 3-year period (2008-2010). Results show that the developed models performed well for geosmin (R (2) = 0.83 for all training data and R (2) = 0.78 for validation of all 10 data points in the validation dataset) and reasonably well for the 2-MIB (R (2) = 0.82 for all training data and R (2) = 0.70 for 7 out of 10 data points in the validation dataset).