We present a new algorithm for multiple approximate string matching. It is based on reading backwards enough l-grams from text windows so as to prove that no occurrence can contain the part of the window read, and then ...