In this paper， Ningxia Yanchi Tan sheep was used as the material， and the LF-NMR detection technology was used to analyze the moisture distribution and migration laws of the mutton during the roasting process at three roasting temperatures. The results showed that the moisture content of mutton was gradually reduced at the three roasting temperatures， and the moisture content was about 55% after roasting for 20 minutes. The L* value of mutton increased at the beginning of roasting process and then decreased with the extension of the roasting time， the a* value showed a first decline and then tended to balance， and the b* value increased at the beginning of roasting process and then tended to balance. At the same time， the combined water content of the three roasting temperatures showed a gradual upward trend， and the non-flowing water content all showed a downward trend. The free water content did not significantly differ during the roasting process (P<0.05). The non-flowing water and free water migration corresponding to the relaxation time of mutton under different roasting times were obvious， indicating that the fluidity of water becomes smaller， and the water with high degree of freedom gradually transfers to the water with low degree of freedom. The brightness of the hydrogen proton density image of mutton at the three roasting temperatures showed a decreasing trend with the extension of roasting time， and the roasting temperature of 250 ℃ was the most obvious， indicating that its internal moisture content decreased the fastest. The total area of low-field nuclear magnetic resonance peaks and the moisture content of mutton in the roasting process showed a significant linear relationship(P<0.05). Sensory evaluation found that roasting for 12， 14 and 18 minutes can be used as the best flavor time for mutton roasting at 250， 220 and 190 ℃， respectively.