Laser-induced breakdown spectroscopy (LIBS) is used to detect chromium content in soybean oil. A series of soybean oil samples with different chromium concentrations are used, and an AvaSpec two-channel spectrometer is used to acquire spectra of samples in the wavelength range of 206.28~481.77 nm. According to the LIBS spectra, several primary characteristic spectral lines of the Cr element are confirmed, then linear regression or least squares support vector machine (LS-SVM) method is used to develop univariate, bivariate and multivariate calibration models. Cr content of the samples is predicted by these calibration models. The results indicate that the performance of bivariate and multivariate calibration models is superior to that of the univariate calibration model, and the performance of the multivariate calibration model developed by LS-SVM is the best. The average relative error (RE) of sample prediction results in univariate and bivariate calibration models is 14.16% and 11.58%, respectively. The average RE of sample prediction in multivariate calibration models developed by linear regression and LS-SVM is 10.95% and 4.97%, respectively. According to these results, the LIBS technique has some feasibility to detect Cr content in soybean oil, and the LS-SVM method can improve the prediction accuracy of calibration models effectively.