Separation of multiple mixing chaotic signals is an important issue in chaos and its applications. An instantaneous blind separation method for linearly mixed chaotic signals is proposed in this paper, in which the uncorrelation characteristics of chaotic signals are utilized. In the case of unknown mixture matrix and chaotic equations, the inverse matrix for reconstructing the source chaotic signals can be directly estimated from the observation by the approach of solving eigenvectors. The results by computer simulation indicate that the multiple mixing chaotic signals, by using the method, can be effectively separated from noisy background even when the signal to noise ration is low.