Abstract: For global navigation satellite system (GNSS) carrier phase relative positioning, the baseline vector can be filtered in each dimension to decrease the noise in static situation. However, the baseline vector may change in dynamic situation, so the filtering method cannot be used. The baseline length is taken as constant in some situations such as orientation and attitude determination. The noise of averaged baseline length is far less than the noise of the single epoch baseline vector if different baseline lengths are averaged by the previous results. Therefore based on previous baseline length, an algorithm was proposed in this paper to decrease the noise of dynamic baseline vector. A mathematical model was developed to combine the averaged baseline length and current baseline vector,and a weight matrix of the mathematical model was deduced. A theoretical expression was found for the noise decreasing on the basis of variation transmission. Simulation and field tests were carried out to evaluate the performance of the proposed algorithm. The results show that, test result matches with the result of theoretical analysis, the proposed algorithm is easy to be implemented, and can effectively decrease the noise of the baseline vector to 80% approximately without additional hardware resources.