Abstract : Term extraction tools extract candidate terms and annotate their occurrences in the texts. However, not all these occurrences are terminological and, at present, this is still a very challenging issue to distinguish when a candidate term is really used with a termino-logical meaning. The validation of term annotations is presented as a bi-classification model that classifies each term occurrence as a termi-nological or non-terminological occurrence. A context-based hypothesis approach is applied to a training corpus: we assume that the words in the sentence which contains the studied occurrence can be used to build positive and negative hypotheses that are further used to classify unde-termined examples. The method is applied and evaluated on a french corpus in the linguistic domain and we also mention some improvements suggested by a quantitative and qualitative evaluation.