This paper examines the creation of a continuous speech recognition system for Pashto and presents the work as a response to a clear gap in existing research. The study explains why earlier efforts focused on digits or isolated words and links this limitation to the lack of a continuous speech database and a phonetic dictionary. To the best of author’s knowledge, no earlier research has been undertaken on the continuous voice recognition of Pashto. This might be due to the unavailability of the Pashto continuous speech database and the Pashto phonetic dictionary. This paper includes the development of a medium size corpus along with a phonetic dictionary that supports practical training needs. Feature extraction utilized Mel Frequency Cepstral Coefficients and Perceptual Linear Prediction features, while classification employed a Hidden Markov Model method, adhering to accepted practices in speech processing and accommodating Pashto phonology. The system scored a recognition rate of 74.25 percent on broadcast speech, a number that shows functional competence given the variability prevalent in such audio.