It is disclosed a blood glucose prediction system and method using saliva-based artificial intelligence deep learning technique. According to one example embodiment, a postprandial blood glucose prediction system may comprise a learning modeling unit for learning a glucose change inference model to infer a pattern difference between blood glucose change and salivary glucose change according to eating by considering a physical indicator, and learning a postprandial blood glucose inference model to infer a correlation between postprandial salivary glucose and postprandial blood glucose by considering the pattern difference; a target information acquiring unit for acquiring a physical indicator and postprandial salivary glucose of a target; a pattern difference estimating unit for estimating a pattern difference of the target by using the glucose change inference model with the physical indicator of the target as an input parameter; and a postprandial blood glucose predicting unit for predicting postprandial blood glucose of the target by using the postprandial blood glucose inference model with the postprandial salivary glucose and the estimated pattern difference of the target as an input parameter.