In this paper we present results of our analysis of isolated utterances of Hindi (an Indian language) stop consonants and propose a recognition method for the same. Standard techniques which use fixed strategies for isolated word speech recognition fail to work for this confusable vocabulary set, since they do not focus on the primarily distinguishing parts of the utterance which contain linguistically important information. The duration of the consonantal part being small, their features are submerged by the dominant vowel part of the utterance. The studies on consonants with different manners of production suggest that processing may be confined to different regions to enhance the recognition performance. An improved performance is obtained by using a hierarchical approach for recognition along with the use of appropriate knowledge for processing different regions of an utterance.