A neural network based estimator designed to be used with the aid of a transformed eigenstructure is proposed for source number estimation in array signal processing. By introducing a similarity transformation, the estimator is utilized to determine the number of sources while overcoming a lack of data samples and noise model information. Unlike conventional methods such as Akaike Information Criterion (AIC) [1] and Minimum Descriptive Length criterion (MDL) [6], the proposed method, called the Neural Fuzzy Estimator (NFE), provides accurate and superior estimation of the source number in simulated data situations of colored noise and limited data.