The purpose of this thesis is to propose the scanning n-tuple (sn-tuple) for on-line handwritten Thai character recognition from word script. The coordinates of characters were chain-coded to convert them to strings. Sn-tuple was then applied to build a statistical model for strings of each character class. Maximum-likelihood was used to classify the characters. To solve the wrong recognition problem of similar characters, condition checking was used i.e. the height and the baseline of the characters, the width of the characters, the ratio of the characters width to their height, the distance between the pen down point and the maximum point, and the consideration of the region. In the postprocessing, maximum score matching was used to recognize each word. The system was executed on microcomputer of Pentium II 400 MHz and 128 Mbytes of RAM. Total single characters used were 10,365 characters from 20 persons. The recognition rate achieved 86.39%. The result of the script of 1,820 words collected from 91 written words per person was 99.67% in the first rank and 100% in the top-3 rank character recognition. The average speed in training was about 380 characters per second and in testing was about 23 characters per second.