Predictive models based on broad scale, spatial surveys typically identify nutrients and climate as the most important predictors of cyanobacteria abundance; however these models generally have low predictive power because at smaller geographic scales numerous other factors may be equally or more important. At the lake level, for example, the ability to forecast cyanobacteria dominance is of tremendous value to lake managers as they can use such models to communicate exposure risks associated with recreational and drinking water use, and possible exposure to algal toxins, in advance of bloom occurrence. We used detailed algal, limnological and meteorological data from two temperate lakes in south-central Ontario, Canada to determine the factors that are closely linked to cyanobacteria dominance, and to develop easy to use models to forecast cyanobacteria biovolume. For Brandy Lake (BL), the strongest and most parsimonious model for forecasting % cyanobacteria biovolume (% CB) included water column stability, hypolimnetic TP, and % cyanobacteria biovolume two weeks prior. For Three Mile Lake (TML), the best model for forecasting % CB included water column stability, hypolimnetic TP concentration, and 7-d mean wind speed. The models for forecasting % CB in BL and TML are fundamentally different in their lag periods (BL=lag 1 model and TML=lag 2 model) and in some predictor variables despite the close proximity of the study lakes. We speculate that three main factors (nutrient concentrations, water transparency and lake morphometry) may have contributed to differences in the models developed, and may account for variation observed in models derived from large spatial surveys. Our results illustrate that while forecast models can be developed to determine when cyanobacteria will dominate within two temperate lakes, the models require detailed, lake-specific calibration to be effective as risk-management tools.