This paper presents a method for fine analysis of children handwriting on pen-based tablets. This work is in the context of the P2IA project, funded by the French government, which aims at designing a virtual notebook in order to foster handwriting learning for primary school pupils. In this work, we consider the task of analysing handwritten words in the context of a dictation exercise. This task is complex due to different factors: the children do not master yet the morphological aspects of handwriting, nor do they master orthography or translating phonetic sounds to actual graphemes (parts of word). In order to tackle this problem, we extend to the context of dictation exercises an analysis engine that was developed previously to deal with copying exercises. Two strategies were developed, the first one is a baseline approach and relies on double child input: the pupil types the word on a virtual keyboard after writing it with the stylus, thus the prior knowledge of the written word will drive the engine analysis. The second one relies on a single input: the child handwritten strokes. To drive the analysis, the strategy consists in generating hypotheses that are phonetically similar to the dictated instruction, which will act as probable approximations of the written word (sequence of letters), to cover potential orthographic mistakes by the pupil. To assist the learning process of the pupils, the engine returns different types of real-time feedbacks, that depend on the confidence of the analysis process (confident assessment on errors, warning, or reject).