In this paper, we present a novel adaptively automated fingerprint classification scheme, which is computationally efficient and resolves both intra-class diversities and inter-class similarities. Initially, preprocessing of fingerprint images is carried out to enhance the image. For classification based on global shape, directional image is computed. Principal component analysis is employed in first stage for dimensionality reduction and to get feature space that accounts for as much of the total variation as possible. In second stage, self-organizing maps are involved for further dimension reduction and data clustering. We use the kohonen topological map for pattern classification. The learning process takes into account the large intra class diversity and the continuum of fingerprint pattern types. Finally LVQ2 maps the class separated fingerprint images into their respective class, the winner and runnerup neuron are trained in such a way that they take into account the inter-class similarities. Experimental results show that AAFFC achieves an accuracy of around 89% for five-class classification tested on NIST 4 without rejection.