A novel method for recognizing fluorescence spectra of hypertriglyceridemia serum was presented based on principal component analysis and probabilistic neural networks. Firstly, two sorts of fluorescence spectra of normal and hypertriglyceridemia serum were measured at 290 nm and 350 nm excitation. And initial feature vectors were obtained from fluorescence intensities at intervals of 1 nm, 2 nm and 5 nm respectively. Secondly, principal component analysis was used to distill initial feature vectors and establish new sample′s feature vectors according to the cumulate reliabilities (>95%). Finally, the probabilistic neural network was designed. Recognition rates with different smoothing parameter and sampling interval were studied. Results show that recognition rates of the normal and hypertriglyceridemia serum are 95% and 100% respectively, when the sampling internal is 5 nm and the smoothing parameter is in range of 0.26~0.92.