In the U.S., educational facilities consume a largeamount of energy. Model predictive control schemes canimprove the energy efficiency of educational facilities.Accurate and fast prediction of the cooling load isessential to performances of model predictive controlschemes. Although many methods for the cooling loadprediction were proposed, they are not suitable foreducational facilities due to the lack of an efficient wayto reflect the impact of internal activities on the coolingload. After analyzing the characteristics of cooling loadof educational facilities, we proposed to use the day typeinstead of the day of the week as the input for theprediction. Then we constructed a Bayesian Networkmodel based on that. To evaluate how the proposedinputs enhance the cooling load prediction, we alsoimplemented the other Bayesian Network model withinputs recommended by the literature. To assessperformances of those models, we performed a casestudy in which on-site measured cooling load andmeteorological data was used for the training and testing.The results show that the Bayesian Network models cancapture the trend of cooling load even with a limited sizeof training data. Replacing the day of the week by theday type can significantly improve the accuracy ofcooling load prediction for educational facilities.