This paper proposes a theoretically rigorous analysis procedure that compares the position domain and range domain carrier-smoothed-code filters for differential GNSS positioning. Utilizing consistent error covariance formulation, it is shown that filtering in the position domain is, in theory, more advantageous than range domain carrier-smoothed-code filtering. It is also shown that if the visible satellite set does not change during a sufficiently long time interval the performances of position and range domain filters are similar.