Based on strong fluorescence property of oil pollutant in water quality, a fluorescence detection system is constructed, gasoline, diesel, kerosene, engine oil with different concentrations in water quality are prepared to analyze their fluorescence characteristics. In order to identify the three oil with the similar fluorescence peaks, the Principal Component Analysis (PCA) combined with Extension Neural Network (ENN) is proposed, which can reduce input vector and identify similar substances. Compare with the ENN and PCA-BP, the results show that the proposed method can make the iteration number dropped from 265 to 60, make the sum of the squares of dispersion decrease from 0.236 5 to 0.014 5, make recognition efficiency increase from 72.50% to 96.25%, and reach the 10-6 level of recognition precision. The proposed method possesses high recognition accuracy and recognition efficiency, which can be used in spectral identification of other trace organic materials in water.