As an important means of managing and monitoring the electromagnetic spectrum, communication signal modulation recognition shows important research value and application prospects. A signal modulation recognition method based on complex neural network is proposed by using the frequency domain information of modulated signals for modulation recognition. Firstly, the I and Q signals are combined into complex signals, and the real and imaginary parts obtained are combined as the data set of the input network after Fast Fourier Transform(FFT). Secondly, a complex neural network structure is designed, and an attention mechanism is introduced to improve the network structure. Finally, the simulation results show that the proposed method can effectively identify nine modulation modes, and the average correct recognition rate reaches 96.33% when the signal-to-noise ratio is 6 dB.