An algorithm is proposed which is a generalization of the method of least squares, which produces strongly consistent estimators of the parameters of bilinear dynamical systems with noise in the output signal in the absence of information about the distribution of the noise.