A dynamic model of a closed-loop spray cooling system was proposed, and a model predictive temperature control algorithm was presented. The presented algorithm could quantitively estimate sudden changes in heat load, and then plan an optimized control path along which the temperature trajectory was numerically predicted to be the most stable. Then, the presented algorithm was integrated into a closed-loop spray cooling system, and its control performance was numerically studied. The results showed that, following a 20 W downward step change in heat load, the temperature was re-stabilized in 4 s, with its oscillation less than 0.14 K. Under identical conditions, the presented algorithm outperformed PID in terms of temperature stability, with its peak time shorter by 25–55%, settling time shorter by 38–51%, and average temperature offset smaller by 50–60%. This advantage was mainly because the presented algorithm compensated for hydraulic and thermal lags, two sources of control nonlinearity, by excessive regulation in advance. Unexpectedly, temperature overshoots of the presented algorithm and PID did not differ significantly, seemingly due to the algorithm-inherent estimation lag. From a system design perspective, the length of pump pipeline should be as short as possible, which proved to help reduce overshoot and settling time. Similarly, addition in heat load’s mass also helped reduce overshoot, but would result in an unfavorable long settling time, and vice versa. The simulation results were validated by experimental data, with error less than 10%.