Aiming at the defects of the common communication emitter recognition method, that the targets can't be recognized in the condition of low SNR, a modified communication emitter recognition method based on evidence reasoning theory is proposed. Characteristic selection and database reconstruction is applied to solve the problem of indistinct input, and the D-S algorithm of weighted averaging is applied to make inferences about signal characteristics in all levels which can avoid inefficient of data fusion. The simulation results confirm that the new method based on modified D-S theory can observably improve the performance of recognizing communication emitter for low SNR.