An efficiency of two blind signal extraction algorithms (gradient and fixed point) based on the maximization of absolute kurtosis is analyzed in the work. Algorithms allow to extract a pulse signal on the background of impulse noise. The algorithms’ convergence at and nonzero initial conditions is proved. A lemma and two theorems which allow to prove blind allocation signal and to determine the number of decisions regarding the extracting signal are formulated. It is determined that the fixed point algorithm based on maximizing the absolute kurtosis is more efficient. It allows extracting the pulse desired signal having the signal-to-noise ratio 30 dB greater than the gradient algorithm with the same objective function.