In this report we describe how the QALC system (the Question-Answering program of the LIR group at LIMSI-CNRS, already involved in the QA-track evaluation at TREC9), was improved in order to better extract the very answer in selected sentences. The purpose of the main Question-Answering track in TREC10 was to find text sequences no longer than 50 characters or to produce a "no answer" response in case of a lack of answer in the TREC corpus. As QALC first retrieves relevant sentences within the document corpus, our main question was: how to find the answer in a sentence? This question involves two kinds of answer: a) it is better to know what you look for and b) you have to know the location of what you look for. The first case is solved by applying a question analysis process. This process determines the type of the expected answer in term of named entity. However, all answers cannot be expressed in term of a named entity. Definition questions or explanation questions for example demand phrases (noun phrases or verb phrases) as answers. So, after having studied the structure of subpart of sentences that contained answers, we defined criteria to be able to locate the precise answer within a sentence. These criteria consist in defining triplets composed of a question category, the question focus and an associated list of templates allowing the location of the answer according to the focus place in the candidate sentence.