Carrier-phase inter-frequency bias (IFB) exists in GLObal NAvigation Satellite System (GLONASS) baselines when receivers of different brands are used. Those biases need to be calibrated in GLONASS data processing to derive precise fixed solutions. Consequently, the accuracy of the IFB calibration affects ambiguity fixing, and low-accuracy IFB values in the calibration will degrade the positioning results. Hitherto, at least two IFB rate value sets for various receiver brands have been given in previous studies. Some of the differences between the value sets exceed 2 mm/FN (frequency number) and the effects of those differences on ambiguity fixing have not been investigated until now. This study showed that ambiguity fixing in GLONASS single-epoch positioning is very sensitive to the accuracy of the IFB rates. Even errors of millimetres can seriously lower the empirical success rate. The short baselines from the Global Navigation Satellite System (GNSS, which includes Russian GLONASS) networks of the International GNSS Service and the Regional Reference Frame Sub-Commission for Europe were employed to obtain more accurate IFB rate estimates with the proposed two-dimensional particle filtering. Afterwards, the IFB estimates were statistically refined by the least-squares method. Experiments showed that when the refined IFB rates were used in the IFB calibration, the empirical success rates of ambiguity fixing in GLONASS single-epoch positioning were largely improved compared with the values given before, improvements such as 20·6% for a Septentrio and Leica receiver combination.