This paper presents the design and implementation of a fingertip writing interface which recognizes the moving trajectory of the user?s fingertip into alphabets and numerals. The processes are divided into tracking and recognition. For the fingertip tracking process, the interface employees techniques including background subtraction, skincolor modeling, finger extraction, fingertip positioning and Kalman filter prediction. To recognize the fingertip trajectories, four types of features are defined for recognition with Hidden Markov Models. According to our performance evaluation, the writing interface achieves an accuracy rate of 98% for fingertip tracking and reaches a recognition accuracy as high as 93% for alphabets and numerals, demonstrating its potential to serve as a feasible human-machine interface of natural modality.