The invention discloses a soft measurement method for detecting the temperature distribution of billet inside a metallurgy heating furnace. The method selects a plurality of key variables affecting the temperature distribution of billet inside the furnace to form a process detection variable set, applies partial least square technology based on multivariate statistics projection principle to establish a soft measurement model between the temperature distribution variable of billet of the metallurgy heating furnace and the process detection variable and determines the optimal pivot numbers through cross check technology. The method avoids complicated process mechanism analysis, has convenient field implementation and high soft measurement precision and is particularly applicable to such industrial occasions similar to the temperature distribution detection inside the metallurgy heating furnace with high dimension and very rich process data. The soft measurement method can be used for monitoring in real time or directing actual production, improve the product quality, increase the yield and realize the energy conservation and consumption reduction of the metallurgy heating furnace overall.