Regarding the problem of long estimation time and big estimation error of tag estimation algorithms that currently exist in Radio Frequency Identification (RFID) system, a novel tag estimation method based on non-empty slot number was proposed. Firstly, the model of Dynamic Frame Slot ALOHA (DFSA) algorithm was analyzed to point out the necessity of tag estimation; secondly, some tag estimation algorithms were researched and their shortcomings were listed; thirdly, by researching the relationship between the average number of non-empty slots and the number of total tags to be identified in different frame length conditions, a normalized curve independent of frame length was obtained and applied to estimate tag numbers. Furthermore, by introducing accuracy demand, the total number of polling times K under different tag numbers was determined by using probability theory and binary search method; finally, a simulation was conducted and a comparative analysis of performance between the proposed method and some existing tag estimation algorithms was done from aspects of estimation accuracy and estimation time. The simulation results show that the maximum error rate of the proposed algorithm is only 1%. When the frame length was 128 and the total tag number was 400, the error rate of the proposed tag estimation algorithm was reduced by 66.7%, 78.3% and 72.2% respectively compared with Adaptive Slotted ALOHA Protocol (ASAP), Fast Zero Estimation (FZE) and Maximum A Posteriori (MAP) estimation algorithm. What's more, in the case of recognizing the same number of tags, the estimation time of the proposed algorithm was also significantly less than that of the above-mentioned three algorithms. Thus, the proposed method has higher estimation accuracy and estimation efficiency and can identify tags quickly and accurately in RFID system.