For the problem that estimating multiple sinusoidal signal frequency is insufficient by using the classical Prony algorithm under the noise background, this paper proposes an improved Prony algorithm with strong antinoise ability. Through establishing a new accumulation relationship, the proposed algorithm can only use appropriate new sequence values to reconstruct the sparse sums linearly under the condition of low SNR, so as to more accurately estimate signal frequency. It compares the proposed algorithm with the classical Prony algorithm by making the simulation experiment. And the experimental results show that as the signal embedded in the noise, the classical Prony algorithm loses effectiveness for the signal frequency estimation, and while the new algorithm can still effectively estimate signal frequency. So, the proposed algorithm has strong antinoise ability, relative stable performance and high precision advantages and shows better practicability in the multiple sinusoidal signal frequency estimation.