Dinoflagellate bloom consecutively occurred in Caojie Reservoir from 2011 to 2013 and threatened the local economy and ecology.Recently, support vector machine(SVM) was reported to have advantages of only requiring a small amount of samples, high degree of prediction accuracy, and generalization to solve the nonlinear classification problems. In this study, the SVM-based prediction model for dinoflagellate bloom was established by monthly field date collected from May 2011 to July 2013 at 8 transects in Caojie Reservoir. The maximum accuracy excessed 80% by choosing environmental variables data and Peridiniopsis niei abundance of current month, and accuracy arrived at 100% for dinoflagellate bloom samples. The results showed that the SVM classification is an effective new way that can be used in monitoring dinoflagellate bloom in Caojie Reservoir and have a promising application prospect for environmental protection.