Challenges remain in imaging fast biological processes in vivo with fluorescence molecular tomography (FMT) due to the long data acquisition time. Data acquisition with limited projections can greatly reduce the time consumption, but the influence of limited-projection on reconstruction quality is currently unclear. Both numerical simulations and a phantom experiment are conducted to analyze this problem. Through a systematic investigation of all the results reconstructed from different numbers of projections, we evaluate the influence of limited-projection data on FMT. A mouse experiment is also performed to validate our work. A general relationship between the number of projections and reconstruction quality is obtained which indicates that the projection number of three is preferred for fast FMT experiment.