The paper introduces a new model of the so-called Growing Hierarchical Neural Networks (GHNN) applicable to image classification without requiring any special domain-specific pre-processing. Two experimental case studies involve face and hand-written digit recognition.