Abstract: In poultry breeding, chicken gender is closely related to its productivity.For the rapid and nondestructive detection on the gender of chicken embryos in early period during the whole hatching, a laboratory hyperspectral imaging system was setup to capture hyperspectral transmission images of 94 hatching eggs on days 2, 4, 6, 8, 10 and 12, at the spectral region of 400～1 000 nm.On the 16th day during the hatching, chicken egg embryos were broken for gender determination by human visual inspection from three trained experts, due to the fact that the difference in morphology between male and female chicken embryos is enough to be identified by human vision on day 16.After comparisons among the images information between male and female embryos, no clear differences was found.Therefore, the spectral information was further analyzed.The regions of interest (ROIs) were selected at big head, middle part and small head of egg for spectral response extraction.For the optimal time and location selection for gender determination, three models including linear algorithm(partial least squares discriminant analysis, PLSDA) and nonlinear algorithm(support vector machine, SVM; artificial neural network, ANN) were built by using full band spectral response, and the discrimination accuracy for male and female embryos were compared at different time and locations based on different models.The results showed that all of the discrimination models had the best accuracy on the 10th day of incubation as well as the ROI located on the middle of egg.Next, the average spectral response difference between male and female embryos was analyzed.600～900 nm was determined by removal of bands of 400～599 nm and 901～1 000 nm, which had little difference between male and female embryos, to build the gender discrimination models based on SVM, PLSDA and ANN.The results showed that the accuracy of SVM and PLSDA were both 75.00% for prediction, which were better than the accuracy with 71.88% for both SVM and PLSDA, based on full band spectral response on day 10 and middle part.For ANN, the prediction accuracy was 82.86%, by an increase of 2.86% compared with full band spectral response.Furthermore, the overall prediction performance of ANN model had better results for gender detection than SVM and PLSDA.The results in the study showed that hyperspectral imaging technology has the potential to identify the gender of early hatching chicken eggs embryos, but the accuracy of identification may be affected by the individual differences of hatching egg.So, in further research, more samples, physicochemical and biochemical differences between male and female embryos in various developmental stages were suggested to be incorporated for higher accuracy acquision of gender discrimination.