Face Recognition (FR) has witnessed numerous local descriptors in the last few decades. But their performances are not so impressive under illumination variations (challenge). This is due to incomplete methodologies or not effective utilization of pixels are done from lower and higher scale patches. This work introduces a MB-ZZLBP descriptor for face analysis. In MB-ZZLBP, first 9 × 9 block is taken and median values are computed in each region of the 9 × 9 block. In 9 × 9 block there are 9 regions and each block have the size of 3 × 3. After median computation 3 × 3 block evolves which contains the median values. By computing median in blocks the illumination noise is greatly reduced. Further zigzag ordered pixels are compared to form the code of MB-ZZLBP, which further leads 256 size. FLDA with SVMs and NN are taken for compaction and matching. Experiments on EYB, YB, and GT proves developed method efficacy in contrast to others. MB-ZZLBP secures the best ACC rates of 99.69, 100, and 98% on the considered datasets. Further 24 methods are outclassed by the MB-ZZLBP. This shows how emphatic and effective is MB-ZZLBP. The significance of MB-ZZLBP has been justified under challenge illumination variations, by introducing a novel methodology in terms of the novel descriptor MB-ZZLBP. The results of MB-ZZLBP are far better than the various compared methods. This proves the MB-ZZLBP potent.