CMOS chips having high leakage are observed to have high burn-in fallout rate. I-DDQ testing has been considered as an alternative to burn-in. However, increased subthreshold leakage current in deep sub-micron technologies limits the use of I-DDQ testing in its present form. In this work, a statistical outlier rejection technique known as the median of absolute deviations (MAD) is evaluated as a means to screen early failures using I-DDQ data. MAD is compared with delta I-DDQ and current signature methods. The results of the analysis of the SEMATECH data are presented.