Abstract: Heteroscedastic and time correlation characteristics of the GPS observation sequence are the main error structures for GPS stochastic modeling.When coping with such problem,the most difficult thing is that the abovementioned two factors could not be well taken into account at the same time.That is to say,stationary process or widely stationary process is the basic requirement when dealing with the time correlation analysis;and analogously,time independent is also the prerequisite to deal with the variance evaluation.To overcome such problem,a new method is proposed.With this method,the heteroscedastic and time correlation characteristics are needed to be considered in sequence.The experimental result proves that the method is correct;and also it shows that the positioning result and the corresponding reliability are greatly improved by the proposed method.