In this paper, an evaluation of unsupervised Kohonen Self-Organizing Maps (SOM) and Crisp Vector Quantization (CVQ) classifiers for speaker recognition in limited data condition is presented. We demonstrate through experimental studies that fine tuned SOM yields better recognition performance over CVQ especially in limited data conditions. We already have proposed Variable Frame Rate and Size (VFRS) analysis of speech improves the recognition performance compared to Single Frame Size (SFS) analysis in limited data conditions using CVQ. VFRS technique extracts the features using multi-resolution and multi-shifting concepts. Alternatively, in case of SFS technique the features are extracted with single frame size and rate. To show an alternative model still improves the recognition performance in limited data conditions, we conducted this experiment using SOM. Experimental results show that SOM models are better in limited data conditions. Copyright © 2007 IICAI.