Due to the impact of swine fever, the demand for eggs which is an important substitute for pork, has increased significantly, and the laying hens breeding industry has also gradually developed and expanded to meet people's demands. Therefore, it is of great significance for the development of layer breeding industry that how to judge gender at the stage of chick and even embryo development accurately and conveniently. To this, 96 fresh seed eggs with similar shell color and no cracks on the surface were selected, and the visible/near-infrared diffuse reflection spectrum was used as the research object, investigated the influence of data collection location and spectral pretreatment method on the qualitative model of gender identification of seed eggs. Diffuse reflectance spectral intensity was collected at three different positions on the surface: blunt end, sharp end and the equator. After correction, 440.27~874.6 nm was selected as the effective spectral band for analysis. The spectral intensity was calculated according to 2∶1 divided into a training set and test set, the proportion of the normalized (Normalize), the second Derivative (2nd Derivative), standard normal variable transformation (SNV), multiple scatter correction (MSC), to trend method (Detrend), spectral transformation method (Spectroscopic), a total of six kinds of pretreatment of PLS-DA model and LDA models' prediction accuracy were analyzed, then compared with the original data (Raw) prediction accuracy, the changes of accuracy were obtained. Through comprehensive analysis of spectral data collected 216, 240, 264, 288 and 312 h after incubation and egg gender information at different positions, it was found that the pretreatment effect was the best at 288 h after embryo development, and the accuracy of 35 models was effectively improved. The pretreatment effect at 264 h was the worst in the analysis time, and its treatment reduced the accuracy of 19 models. The pretreatment of 312 h reduced the discriminant accuracy of 12 models. Detrend and Spectroscopic, two kinds of pretreatment method, could significantly improve the effect of discrimination, but the Spectroscopic model may not be able to predict; SNV and MSC had the same effect on the model, Normalize's effect on the model could not be determined. The accuracy of 2nd Derivative treatment was uncertain, which is sometimes consistent with the effect of original data modeling. The comprehensive experimental results showed that the preprocessing used LDA model of 288 h embryo development data could effectively improve the discriminant accuracy of the model, among which the Detrend preprocessing of the data at the blunt end of the egg was good. The results provided a reference for establishing an early and rapid detection model based on visible/near-infrared gender information in egg species.