The purpose of this paper is to examine the ability of different statistical features obtained from denoised vibration signals to distinguish between healthy and defective Rolling Element Bearings (REB) at the six sigma significance level or Defects Per Million Observations (DPMO). Raw vibration signals from the experimental setup were subjected to Interval Dependent denoising. Discrete Wavelet Transforms (DWT) technique was used to generate 17 statistical features for 4 independent conditions of REB, viz., healthy(N), and REB with defects on Inner Race (IR), Outer Race (OR) and Ball (B).  These statistical features were compared using the Independent Samples test at a six sigma significance level. Five statistical features that could distinguish between a defective and a healthy REB viz, Root Mean Square (RMS), Standard Deviation (SD), Square of Mean of Square Root (SMSR), Mean Absolute Value (MAV) and Log-log Ratio (LLR) at six sigma significance were identified. The methodology used in this study is a unique combination of vibration signal analysis, statistical feature extraction and simple inferential analysis of REB defects at very low significance levels comparable with six sigma DPMO. The identified statistical features can be used before predictive analysis.