This paper deals with the automated categorization of on-line handwritten documents. We experimentally show the effects of word recognition errors on a categorization engine using machine learning algorithms. We compared the performances of a categorization system over the texts obtained through on-line handwriting recognition and the same texts available as ground truth. Results show that no significant accuracy loss is expected when about 78% percent of indexation terms are correctly recognized. Results also show that using the top n recognition-candidates increases categorization rates of texts where more than 50% of indexation terms are incorrectly recognized