In order to solve the problem of quantitative description for molten steel flow characteristics in multi-strand tundish, the shortage of two traditional analytical models for flowing character multi-strand tundish is investigated in detail. While the number of the tundish strand is more, the parallel model used for flow characteristics of multi-strand tundish occur the phenomenon of the actual residence time is greater than the theoretical residence time. Determining the minimum response time is difficult point for the whole analysis approach studying the flowing character in multi-strand tundish. The existing processing methods for the minimum response induces the calculating plug volume fraction lower than actual one. Meanwhile, based on the hypothesis of combined model, the residence time distribution function for flow characteristics of molten steel in tundish is analyzed. The relationship between the standard variance of the residence time distribution function and the minimum response time is established. Then the standard variance analysis model is proposed to quantitatively describe molten steel flow characterization in tundish. This model adopts the residence time distribution function variance instead of the minimum response time to determine the plug volume fraction of the flow in tundish. Using the average value of the tracer concentration of each strand is equivalent to the overall RTD (Residence Time Distribution) curve. Therefore, the standard deviation of the overall residence time distribution function of the multi-strand tundish is the root mean square average of the one of the residence time distribution function of each strand. In this work, based on the standard variance analysis model, the whole standard variance model is proposed for melton steel flow characterization in the multi-strand tundish. Compared with the traditional multi-strand tundish analysis model, the whole standard variance analysis model used for characterizing the flow fluid in multi-strand tundish not only solves the difficulty of determining the minimum response time for the whole analysis approach, but also avoids the occurrence of negative dead volume with the parallel model. Physical modeling is carried out to study RTD curves in a five-strand tundish, and the reasonability and the excellent feature of the proposed model are verified.