This paper proposes an efficient classification system for separating normal blood cells from the pathological cells. The suggested system employs an adaptive histogram equalization scheme to reduce the noise present in the microscopic images. Two-dimensional discrete wavelet transform (2D-DWT) is applied separately to the nucleus and cytoplasm region to generate the feature matrix. The significant and uncorrelated features are chosen using a combination of PCA and Bhattacharyya distance. Subsequently, the reduced feature set is fed to the back propagation neural network for classification purpose. A public dataset ALL-IDB1 is used to validate the proposed scheme. It can be seen that the proposed methodology has a better result as compared to its competent schemes. The accuracy of the suggested scheme is found to be 97.11% in case of combined features from nucleus and cytoplasm region whereas the same is found to be 95.19% and 90.38% if the features are taken separately.