It is difficult to overcome the high correlation of double differential observation using double-differenced model and the initialization procedure costs a long time with makes it difficult to apply in practice.To deal this problem,a recursive algorithm based on the single-differenced model is presented in which the observations are not correlated.The new method increases observation information with the aid of code for original carrier phase observation.At the same time the float ambiguity solutions and its variance-covariance matrix can be obtained recursively and it makes the LAMBDA method be used to estimate the integer ambiguities.The method is also less computation with numerical stability.The experiment results show that the time of initialization is reduced to about 1/50 of that using the double-differenced method,so this method can be effectively applied in practical attitude determination for shortening initialization procedure.