Abstract: To improve the accuracy of the estimated saturation flow rate (SFR) at signalized intersection, a method considering interaction between adjustment factors was proposed based on the current SFR model. Considering two factors (percentage of heavy vehicles and lane width), the data on 35 through lanes at 22 signalized intersections were collected. First, a two-way ANOVA method was used to verify the interactions and quantify how much two factors relate. Then, the multiple linear regression (MLR) built the interaction models (considered lane width and heavy vehicle). Finally, with collecting 6 through lanes at 3 signalized intersections in Beijing, the performance and applicability of proposed models were analyzed. Results show that the errors of SFR prediction models with considering interactions are below 15%. They are significantly smaller than the Highway Capacity Manual model error 31.50%, GB50647-2011 model error 28.49%, respectively. The models with interaction terms (error of 11.68% and 11.67%, respectively) are higher than that without interaction (error of 12.52%). Different forms of interaction terms have less impact on model performance. The relationship between lane width factor and heavy vehicles factor is not independent. The accuracy of the estimated SFR is improved by considering the interaction. This is contributed to model localization.