The invention discloses an improved variable step size equivariant adaptive blind source separation method. A new nonlinear function requiring no change of parameter value is adopted in a separation matrix iteration updating formula of the traditional equivariant adaptive blind source separation method, the step size in the separation matrix iteration updating formula is optimized by the chicken swarm optimization (CSO), so that the step size continuously changes, whitening processing is carried out on an observation signal obtained from a source signal by a non-singular matrix A to obtain a preprocessed signal, and an optimal separation matrix is obtained by the improved separation matrix iteration updating formula according to the obtained preprocessed signal to separate mixed signals. By adoption of the method disclosed by the invention, the conflict between a convergence speed and a steady state error is solved, the operating rate is improved, the algorithm complexity is simplified, and thus the method has a wide application prospect in the fields of signal processing, wireless communication, etc.