Identification of the five typical types of white blood cells provides an assay for diagnosis of various diseases. Many automated image processing methods have been proposed to save medical operators from time-consuming and inefficient work, but these two-dimensional image-based methods depend only on the shapes to classify WBCs, of which the results are unreliable and incomplete because there are often many variances in the shapes of the five kinds of WBCs. To overcome these limitations, we proposed a spectral- and morphology-based method for white blood cell segmentation and classification that fully utilizes the abundant information carried in hyperspectral images. Utilization of spectra information makes the classification more robust and efficient. By tuning the coefficient of the spatial kernel through iterative optimization, we achieved enhanced classification accuracy compared with methods where only single features were taken into account. The experimental results demonstrated the high accuracy of this method for classifying white blood cells into the five types where the overall accuracy was over 90%. However, the spectral range of the hyperspectral imaging system is from 550 nm to 1000 nm which may not cover the whole spectral signatures of each kind of white blood cell. In the future study, we will update the hardware of the hyperspectral imaging system to obtain a broader spectral range, which will be helpful for investigating the distinct spectrum, the intrinsic relationship between the blood cell components and their spectrum differences. Then, more experiments can be tested on the connection between typical spectral bands and the biological samples. Meanwhile, we will also conduct experiments using diseased blood to verify the correctness of our proposed method. In addition, the hyperspectral blood image samples in the present study contained only one white blood cell in each image. We plan to expand our work to examine more complicated samples with multiple white blood cells mixed in one image.