In order to assess quantitatively the prediction properties of the variable length Markov model, we used a test text and predict each letters using the sequence of previous letters. So, a sequence corresponding to the beginning of the test text is given to the VLMM and the prediction of the next letter is compared to the next letter in the test text. Then, we count the number of correctly predicted letters and divide result by the number of prediction done. So we obtain a percentage of good predictions. The two next sections present these results.