The paper presents a stochastic model predictive control (SMPC) framework for buildings with mixed-mode cooling and demonstrates a comparison with deterministic model predictive control (DMPC) and standard rule-based control (RBC). In this study, a probabilistic model of occupants' behavior on window blind closing event is used to represent the stochastic disturbance acting on the system over the prediction horizon. Monte Carlo (MC) simulation was used to capture this stochastic effect. Simulation results show that SMPC may lead to higher amount of energy consumption, thereby providing a more realistic evaluation for the performance bounds of predictive control in mixed-mode buildings since it considers the occupant-building interactions. It was found that SMPC results in lower thermal comfort violations than DMPC and significantly lower compared to rule-based control strategies.