To predict the repast number accurately can reduce the cost of school canteen and improve students' satisfaction. A novel method based on Markov model to predict repast number is proposed according to the consumption situation of campus card system. Firstly, an initial probability is obtained by calculating the eating behavior of breakfast. Secondly, two transfer probability matrices are computed, one is the transfer probability between the behaviors of students having breakfast and having lunch; the other is the transfer probability between the behaviors of students having lunch and having supper. Finally, a Markov model is constructed according to the initial probability and the two probability transfer matrices to forecast the number of diners. The average prediction error of the proposed method is 1.31%, which has a good prediction performance. The experimental results show that the proposed Markov method can capture the students' dining behavior accurately. It may provide valuable information for the school logistics department, contribute to the construction and management of school and meet the needs of students better.