In this paper, an automatic classification method of stellar spectra using the Self-Organization Feature Mapping (SOFM) method is given. The SOFM is an unsupervised learning algorithm of Artificial Neural Network (ANN). It allows the data organized onto a feature graph while conserving most of the topological characters of the original data space. We used this method to classify stellar spectra automatically. The result is very similar to the Harvard sequence with an accuracy comparable to that obtained by human experts. The SOFM should be a useful method for on-line classification of stellar spectrum samples of very large size. The SOFM can be executed automatically, so it can be applied to process large numbers of target spectra.