Abstract: The activity of superoxide dismutase(SOD)in sheep muscle cells was detected by visible microscopic hyperspectral technique. 223 microscopic images were collected by micro-hyperspectral system(380~980 nm). The region of interest was extracted based on the reflectance of sample spectra and the SOD activity was used to establish the model. With the original spectrum combined with a partial least squares regression model,sample set partition and multi-spectral pretreatment models were compared and analyzed. Multivariate scatter correction was preferred as a pretreatment method. Six characteristic methods were used to extract the characteristic wavelengths,and three models of partial least squares regression,multiple linear regression and principal component regression were established according to the characteristic wavelengths. The results showed that the multivariate linear regression model based on competitive adaptive weighting method for selecting characteristic wavelengths was the best. The correlation coefficient and root mean square error of the prediction set were 0.8351 and 21.3578 U/mg·prot.It was feasible to detect superoxide dismutase activity in muscle cells by microscopic hyperspectral imaging.