In order to crack down on camellia oil adulteration and protect the legitimate interests of consumers, near infrared (NIR) spectroscopy and subwindow permutation analysis (SPA) are used to detect complex adulteration of camellia oil (adulterated with soybean oil, colza oil, peanut oil and mixed oil). The NIR spectra of 85 pure camellia oil samples and 315 adulterated camellia oil samples are acquired, and the SPA method is used to select important wavelength variables. After that, partial least squares-linear discriminant analysis (PLS-LDA) is used to develop calibration models for adulteration detection of camellia oil, and the result is compared to the models developed by PLS-LDA with other selection methods such as competitive adaptive reweighted sampling (CARS) and uninformative variable elimination (UVE). The results indicate that NIR spectroscopy combined with SPA method is feasible for complex adulteration detection of camellia oil, the error rate, sensitivity and specificity of classification in prediction set are 0, 1 and 1, respectively. SPA method is superior to UVE method, and is comparable with CARS method. It is an effective variable selection method, and can simplify model, improve model prediction precision and stability.