Speaker Recognition (SR) is an economic method of biometrics because of availability of low cost and high power computers. An important question which must be answered for the SR system is how well the system resists the effects of determined mimics such as those based on physiological characteristics especially identical twins or triplets. In this paper, a new data fusion technique (viz., majority rule for combining evidence from different feature sets) and a new performance measure is proposed for speaker identification of twins in an Indian language, viz., Marathi. The results have been compared with baseline SR system designed by using Linear Prediction Coefficients (LPC), Linear Prediction Cepstral Coefficients (LPCC) and Mel Frequency Cepstral Coefficients (MFCC) as input feature vectors and polynomial classifiers of 2nd and 3rd order approximation for speaker modeling.