According to the characteristics of Latin-Uighur, this paper analyzed the common spelling error types of Latin-Uighur, and then proposed a method which merged the minimum edit distance, directed graph model based lexical segmentation, trigram language model together. Finally, we implemented the automatically spelling check system of Latin-Uighur based on context. It has increased the accuracy of Latin-Uighur spelling check largely. The experiment on the Uighur corpus provided by Xinjiang University reaches an accuracy of 90.1%.