Deep CNNs, though have achieved the state of the art performance in image\nclassification tasks, remain a black-box to a human using them. There is a\ngrowing interest in explaining the working of these deep models to improve\ntheir trustworthiness. In this paper, we introduce a Posthoc\nArchitecture-agnostic Concept Extractor (PACE) that automatically extracts\nsmaller sub-regions of the image called concepts relevant to the black-box\nprediction. PACE tightly integrates the faithfulness of the explanatory\nframework to the black-box model. To the best of our knowledge, this is the\nfirst work that extracts class-specific discriminative concepts in a posthoc\nmanner automatically. The PACE framework is used to generate explanations for\ntwo different CNN architectures trained for classifying the AWA2 and\nImagenet-Birds datasets. Extensive human subject experiments are conducted to\nvalidate the human interpretability and consistency of the explanations\nextracted by PACE. The results from these experiments suggest that over 72% of\nthe concepts extracted by PACE are human interpretable.\n