Multispectral thermometry inversion based on the continuous temperature measurement algorithm has attracted more and more attentions because there is no prior assumption of emissivity models, but needs to set the appropriate initial temperature and emissivity range and the iteration time is longer, and not suitable for on-line measurement. An improved continuous temperature method searching emissivity deviation was proposed for fast true temperature inversion. The problem of searching the emissivity in the continuous temperature measurement method was changed to search the emissivity deviation, and the function relationship between the deviation of each iteration result and the deviation of the iteration stop condition and the error of emissivity was fitted. Through this function relationship, the emissivity deviation was determined by each iteration result, which could quickly reduce the search range of the emissivity, and improve the efficiency of the calculation. Simulation results for 4 kinds of spectral emissivity models showed that the new algorithm did not need to set the initial temperature range. compared with the continuous measurement method, the computational efficiency was improved by more than 60% in the premise of ensuring the accuracy of inversion.