This paper presents an extension of our on-line sentence recognition system by integrating an automatic word extraction mechanism. Our word extraction task is based on the characterization of inter-stroke gaps, combined to a rejection strategy to evaluate the reliability of the gap classification results. A reconsideration mechanism then used this confidence index to create additional extracted word hypotheses by further controlling the complexity of the recognition task. Different metrics are used to evaluate the impact of this whole word extraction task on the recognition performance, on a set of 395 English sentences.