Linear regression is used to analyse the NIR spectra (750～2 500 nm) of p-diethylbenzene to determine its purity. This method will greatly reduce the time and expenditure in detecting the sample, and make the synthesizing process of p-diethylbenzene more easy. Compared the results of near-infrared spectra with those of gas chromatogram, it shows that in the wavenumber of 5 835 cm~(-1) the maximum related coefficient (0.992 9) is reached and the average prediction relative error is 2.72%. All results prove the practicability of this method discussed in this paper.