Surface enhanced Raman spectroscopy (SERS) and quick pre-treatment technology were used to detect triazophos residues in flesh of navel orange. Quantitative analysis model was developed by partial least squares (PLS) algorithm. SERS of different concentration (0.5 to 20 mg·L-1) triazophos juice solution with flesh extract as the matrix were collected by laser Raman spectrometer. Three preprocessing methods such as normalization, MSC and SNV were used to optimize Raman signals and PLS models were set up. The results showed that minimum detection concentration for triazophos in navel orange below 0.5　mg·L-1. The model built with normalization pre-processing gave the best result; the values of correlation (Rp) and Root mean square error of prediction set (RMSEP) were 1.38 and 0.976　6, respectively. The predict recoveries were 95.97%～103.18% and the absolute values of relative errors were below 5%. T-test (t=－0.018) showed that there was no significant difference between the true values and prediction values. This study demonstrates that this method is accurate and reliable.