Abstract: Olive oil adulteration is has ever been increasing in China in recent years. In this study, a portable non-destructive testing instrument was developed to detect the extra virgin olive oil adulteration using the visible/near-infrared spectroscopy. The instrument was consisted mainly of a multi-spectral acquisition module, a control and display module, a power supply module, and peripheral circuits. And then, the transmission spectrum of the adulterated sample was collected, according to the characteristics of extra virgin olive oil. Firstly, corn oil, peanut oil, rapeseed oil, and soybean oil from eight origins were used as the adulterated oils, and then mixed with the extra virgin olive oil from three origins in Spain, Italy, and Greece. 339 adulterated extra virgin olive oil samples were prepared with the mass fractions of 0, 1%, 3%, 6%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, and 60%, respectively. Secondly, a self-built spectral acquisition system was used to collect transmission spectra of the samples. Finally, partial least squares regression, support vector regression, and error back-propagation neural network models were used to analyze the feasibility of adulteration detection. The results show that the correlation coefficients of the calibration set were between 0.947 and 0.987 after spectral preprocessing. RMSEC values were between 0.031 and 0.063. The correlation coefficients of the validation set were between 0.956 and 0.990, and RMSEP values were between 0.026 and 0.063. The characteristic wavelengths for olive oil adulteration detection were extracted using the successive projection. A multi-channel spectral sensor was finally selected to contain 18 bands. A portable and low-cost olive oil adulteration detection instrument was designed, according to the physical characteristics of extra virgin olive oil. The spectral data of adulterated oil was collected by the R&D device. The quantitative analysis models were established for the olive oil adulteration rate using partial least squares regression, support vector regression, and error back propagation neural network. The prediction models were compared as well. The support vector regression model shared the best fit and optimal prediction, with the correction and validation set coefficients of determination of 0.989 and 0.965, and root mean square errors of 0.020 and 0.037, respectively. The prediction model with the support vector regression models was loaded into the detection device, in order to verify the accuracy and stability of the model and detection device. Within the range of 0-60% adulteration rate, three types of extra virgin olive oil adulterated samples were failed to coincide with the concentration of the model training set samples, which were randomly configured every 10%. The stability of the device was characterized by the coefficient of variation in the three predicted adulteration rates of the extra virgin olive oil. Within the range of adulteration rates, the average coefficient of variation was 0.019 in the adulterated samples, indicating the better adaptability and consistency of the model. The testing instrument showed the a highly stable performance with little fluctuation in the prediction over multiple tests. The accuracy of the device was verified to analyze the residuals between the predicted and the true values of the adulterated samples. The maximum absolute value was 4.43 for the residuals of the adulterated samples. The stability and accuracy of the portable device were achieved to analyze the adulteration rate of extra virgin olive oil, fully meeting the needs of on-site testing.