This study examines an enabling condition for natural language access to medical knowledge resources (Medline, CISMeF) indexed with controlled vocabularies (e.g., the MeSH): is the vocabulary of user queries comparable with that of the index terms? The two vocabularies were compared in their original form, then under incrementally normalized forms, using character-based normalizations then linguistic normalizations. Only 16.7% of the user vocabulary, in its original form, is in the MeSH. Progressive normalizations increase this proportion to 65.5%. Besides, if the frequencies of occurrence of words are taken into account, 89.3 % of user word occurrences can be matched to MeSH words. This shows the interest of taking into account further matching methods between queries and index terms than those presented here.