In the present study, a new approach to fast determining the content of urea, biuret and moisture in compound fertilizer composed of urea, ammonium dihydrogenphosphate and potassium chloride was proposed by using near infrared diffuse reflectance spectroscopy. After preprocessing the original spectrum, partial least squares (PLS) models of urea, biuret and moisture were built with the R2 values of 0.986 1, 0.977 0 and 0.971 3 respectively, the root mean square errors of cross validation were 2.59, 0.38, 0.132 respectively. And the prediction correlation factors were 0.973 3, 0.921 5 and 0.967 9 respectively. The authors detected six kinds of compound fertilizer in market for the model verification, the correlation factors were 0.923 7, 0.978 6 and 0.987 4 respectively. The data implied that the new method can be used for situ quality control in the production process of compound fertilizer.