To recover the initial conditions of coupled map lattices (CML), which goes athwart the coupled systems, is one of key tasks of signal processing. Based on thecommon symbolic dynamic method, a novel time-varying symbolic dynamic method is proposed. The Logistic map is used to test the properties of the method. Numerical simulation shows that the new scheme can recover the statistical property of the initial conditions from coupled map lattices for both noiseless and noisy signals. The performance of the proposed method is significantly superior to the common method. Moreover, the proposed algorithm has small biases, small mean square errors and a high correlation for the given data. Our new approach is more appropriate to the analysis of practical systems and measurement of coupled systems.